file.c 76.2 KB
Newer Older
C
Chris Mason 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 * Copyright (C) 2007 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

C
Chris Mason 已提交
19 20 21 22 23 24 25 26
#include <linux/fs.h>
#include <linux/pagemap.h>
#include <linux/highmem.h>
#include <linux/time.h>
#include <linux/init.h>
#include <linux/string.h>
#include <linux/backing-dev.h>
#include <linux/mpage.h>
27
#include <linux/falloc.h>
C
Chris Mason 已提交
28 29 30 31
#include <linux/swap.h>
#include <linux/writeback.h>
#include <linux/statfs.h>
#include <linux/compat.h>
32
#include <linux/slab.h>
33
#include <linux/btrfs.h>
34
#include <linux/uio.h>
C
Chris Mason 已提交
35 36 37 38 39
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
#include "btrfs_inode.h"
#include "print-tree.h"
40 41
#include "tree-log.h"
#include "locking.h"
J
Josef Bacik 已提交
42
#include "volumes.h"
J
Josef Bacik 已提交
43
#include "qgroup.h"
C
Chris Mason 已提交
44

45
static struct kmem_cache *btrfs_inode_defrag_cachep;
C
Chris Mason 已提交
46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70
/*
 * when auto defrag is enabled we
 * queue up these defrag structs to remember which
 * inodes need defragging passes
 */
struct inode_defrag {
	struct rb_node rb_node;
	/* objectid */
	u64 ino;
	/*
	 * transid where the defrag was added, we search for
	 * extents newer than this
	 */
	u64 transid;

	/* root objectid */
	u64 root;

	/* last offset we were able to defrag */
	u64 last_offset;

	/* if we've wrapped around back to zero once already */
	int cycled;
};

71 72 73 74 75 76 77 78 79 80 81 82 83 84 85
static int __compare_inode_defrag(struct inode_defrag *defrag1,
				  struct inode_defrag *defrag2)
{
	if (defrag1->root > defrag2->root)
		return 1;
	else if (defrag1->root < defrag2->root)
		return -1;
	else if (defrag1->ino > defrag2->ino)
		return 1;
	else if (defrag1->ino < defrag2->ino)
		return -1;
	else
		return 0;
}

C
Chris Mason 已提交
86 87 88 89 90 91 92 93 94
/* pop a record for an inode into the defrag tree.  The lock
 * must be held already
 *
 * If you're inserting a record for an older transid than an
 * existing record, the transid already in the tree is lowered
 *
 * If an existing record is found the defrag item you
 * pass in is freed
 */
95
static int __btrfs_add_inode_defrag(struct inode *inode,
C
Chris Mason 已提交
96 97 98 99 100 101
				    struct inode_defrag *defrag)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct inode_defrag *entry;
	struct rb_node **p;
	struct rb_node *parent = NULL;
102
	int ret;
C
Chris Mason 已提交
103 104 105 106 107 108

	p = &root->fs_info->defrag_inodes.rb_node;
	while (*p) {
		parent = *p;
		entry = rb_entry(parent, struct inode_defrag, rb_node);

109 110
		ret = __compare_inode_defrag(defrag, entry);
		if (ret < 0)
C
Chris Mason 已提交
111
			p = &parent->rb_left;
112
		else if (ret > 0)
C
Chris Mason 已提交
113 114 115 116 117 118 119 120 121 122
			p = &parent->rb_right;
		else {
			/* if we're reinserting an entry for
			 * an old defrag run, make sure to
			 * lower the transid of our existing record
			 */
			if (defrag->transid < entry->transid)
				entry->transid = defrag->transid;
			if (defrag->last_offset > entry->last_offset)
				entry->last_offset = defrag->last_offset;
123
			return -EEXIST;
C
Chris Mason 已提交
124 125
		}
	}
126
	set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
C
Chris Mason 已提交
127 128
	rb_link_node(&defrag->rb_node, parent, p);
	rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
129 130
	return 0;
}
C
Chris Mason 已提交
131

132 133 134 135 136 137 138
static inline int __need_auto_defrag(struct btrfs_root *root)
{
	if (!btrfs_test_opt(root, AUTO_DEFRAG))
		return 0;

	if (btrfs_fs_closing(root->fs_info))
		return 0;
C
Chris Mason 已提交
139

140
	return 1;
C
Chris Mason 已提交
141 142 143 144 145 146 147 148 149 150 151 152
}

/*
 * insert a defrag record for this inode if auto defrag is
 * enabled
 */
int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
			   struct inode *inode)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct inode_defrag *defrag;
	u64 transid;
153
	int ret;
C
Chris Mason 已提交
154

155
	if (!__need_auto_defrag(root))
C
Chris Mason 已提交
156 157
		return 0;

158
	if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
C
Chris Mason 已提交
159 160 161 162 163 164 165
		return 0;

	if (trans)
		transid = trans->transid;
	else
		transid = BTRFS_I(inode)->root->last_trans;

166
	defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
C
Chris Mason 已提交
167 168 169
	if (!defrag)
		return -ENOMEM;

170
	defrag->ino = btrfs_ino(inode);
C
Chris Mason 已提交
171 172 173 174
	defrag->transid = transid;
	defrag->root = root->root_key.objectid;

	spin_lock(&root->fs_info->defrag_inodes_lock);
175 176 177 178 179 180 181 182 183 184
	if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
		/*
		 * If we set IN_DEFRAG flag and evict the inode from memory,
		 * and then re-read this inode, this new inode doesn't have
		 * IN_DEFRAG flag. At the case, we may find the existed defrag.
		 */
		ret = __btrfs_add_inode_defrag(inode, defrag);
		if (ret)
			kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
	} else {
185
		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
186
	}
C
Chris Mason 已提交
187
	spin_unlock(&root->fs_info->defrag_inodes_lock);
188
	return 0;
C
Chris Mason 已提交
189 190 191
}

/*
192 193 194
 * Requeue the defrag object. If there is a defrag object that points to
 * the same inode in the tree, we will merge them together (by
 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
C
Chris Mason 已提交
195
 */
196 197
static void btrfs_requeue_inode_defrag(struct inode *inode,
				       struct inode_defrag *defrag)
198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	int ret;

	if (!__need_auto_defrag(root))
		goto out;

	/*
	 * Here we don't check the IN_DEFRAG flag, because we need merge
	 * them together.
	 */
	spin_lock(&root->fs_info->defrag_inodes_lock);
	ret = __btrfs_add_inode_defrag(inode, defrag);
	spin_unlock(&root->fs_info->defrag_inodes_lock);
	if (ret)
		goto out;
	return;
out:
	kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
}

C
Chris Mason 已提交
219
/*
220 221
 * pick the defragable inode that we want, if it doesn't exist, we will get
 * the next one.
C
Chris Mason 已提交
222
 */
223 224
static struct inode_defrag *
btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
C
Chris Mason 已提交
225 226
{
	struct inode_defrag *entry = NULL;
227
	struct inode_defrag tmp;
C
Chris Mason 已提交
228 229
	struct rb_node *p;
	struct rb_node *parent = NULL;
230 231 232 233
	int ret;

	tmp.ino = ino;
	tmp.root = root;
C
Chris Mason 已提交
234

235 236
	spin_lock(&fs_info->defrag_inodes_lock);
	p = fs_info->defrag_inodes.rb_node;
C
Chris Mason 已提交
237 238 239 240
	while (p) {
		parent = p;
		entry = rb_entry(parent, struct inode_defrag, rb_node);

241 242
		ret = __compare_inode_defrag(&tmp, entry);
		if (ret < 0)
C
Chris Mason 已提交
243
			p = parent->rb_left;
244
		else if (ret > 0)
C
Chris Mason 已提交
245 246
			p = parent->rb_right;
		else
247
			goto out;
C
Chris Mason 已提交
248 249
	}

250 251 252
	if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
		parent = rb_next(parent);
		if (parent)
C
Chris Mason 已提交
253
			entry = rb_entry(parent, struct inode_defrag, rb_node);
254 255
		else
			entry = NULL;
C
Chris Mason 已提交
256
	}
257 258 259 260 261
out:
	if (entry)
		rb_erase(parent, &fs_info->defrag_inodes);
	spin_unlock(&fs_info->defrag_inodes_lock);
	return entry;
C
Chris Mason 已提交
262 263
}

264
void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
C
Chris Mason 已提交
265 266
{
	struct inode_defrag *defrag;
267 268 269 270 271 272 273 274 275
	struct rb_node *node;

	spin_lock(&fs_info->defrag_inodes_lock);
	node = rb_first(&fs_info->defrag_inodes);
	while (node) {
		rb_erase(node, &fs_info->defrag_inodes);
		defrag = rb_entry(node, struct inode_defrag, rb_node);
		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);

276
		cond_resched_lock(&fs_info->defrag_inodes_lock);
277 278 279 280 281 282 283 284 285 286 287

		node = rb_first(&fs_info->defrag_inodes);
	}
	spin_unlock(&fs_info->defrag_inodes_lock);
}

#define BTRFS_DEFRAG_BATCH	1024

static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
				    struct inode_defrag *defrag)
{
C
Chris Mason 已提交
288 289 290 291 292
	struct btrfs_root *inode_root;
	struct inode *inode;
	struct btrfs_key key;
	struct btrfs_ioctl_defrag_range_args range;
	int num_defrag;
293 294
	int index;
	int ret;
C
Chris Mason 已提交
295

296 297
	/* get the inode */
	key.objectid = defrag->root;
298
	key.type = BTRFS_ROOT_ITEM_KEY;
299
	key.offset = (u64)-1;
300 301 302

	index = srcu_read_lock(&fs_info->subvol_srcu);

303 304
	inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
	if (IS_ERR(inode_root)) {
305 306 307
		ret = PTR_ERR(inode_root);
		goto cleanup;
	}
308 309

	key.objectid = defrag->ino;
310
	key.type = BTRFS_INODE_ITEM_KEY;
311 312 313
	key.offset = 0;
	inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
	if (IS_ERR(inode)) {
314 315
		ret = PTR_ERR(inode);
		goto cleanup;
316
	}
317
	srcu_read_unlock(&fs_info->subvol_srcu, index);
318 319 320

	/* do a chunk of defrag */
	clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
C
Chris Mason 已提交
321 322
	memset(&range, 0, sizeof(range));
	range.len = (u64)-1;
323
	range.start = defrag->last_offset;
M
Miao Xie 已提交
324 325

	sb_start_write(fs_info->sb);
326 327
	num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
				       BTRFS_DEFRAG_BATCH);
M
Miao Xie 已提交
328
	sb_end_write(fs_info->sb);
329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351
	/*
	 * if we filled the whole defrag batch, there
	 * must be more work to do.  Queue this defrag
	 * again
	 */
	if (num_defrag == BTRFS_DEFRAG_BATCH) {
		defrag->last_offset = range.start;
		btrfs_requeue_inode_defrag(inode, defrag);
	} else if (defrag->last_offset && !defrag->cycled) {
		/*
		 * we didn't fill our defrag batch, but
		 * we didn't start at zero.  Make sure we loop
		 * around to the start of the file.
		 */
		defrag->last_offset = 0;
		defrag->cycled = 1;
		btrfs_requeue_inode_defrag(inode, defrag);
	} else {
		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
	}

	iput(inode);
	return 0;
352 353 354 355
cleanup:
	srcu_read_unlock(&fs_info->subvol_srcu, index);
	kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
	return ret;
356 357 358 359 360 361 362 363 364 365 366
}

/*
 * run through the list of inodes in the FS that need
 * defragging
 */
int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
{
	struct inode_defrag *defrag;
	u64 first_ino = 0;
	u64 root_objectid = 0;
C
Chris Mason 已提交
367 368

	atomic_inc(&fs_info->defrag_running);
369
	while (1) {
M
Miao Xie 已提交
370 371 372 373 374
		/* Pause the auto defragger. */
		if (test_bit(BTRFS_FS_STATE_REMOUNTING,
			     &fs_info->fs_state))
			break;

375 376
		if (!__need_auto_defrag(fs_info->tree_root))
			break;
C
Chris Mason 已提交
377 378

		/* find an inode to defrag */
379 380
		defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
						 first_ino);
C
Chris Mason 已提交
381
		if (!defrag) {
382
			if (root_objectid || first_ino) {
383
				root_objectid = 0;
C
Chris Mason 已提交
384 385 386 387 388 389 390 391
				first_ino = 0;
				continue;
			} else {
				break;
			}
		}

		first_ino = defrag->ino + 1;
392
		root_objectid = defrag->root;
C
Chris Mason 已提交
393

394
		__btrfs_run_defrag_inode(fs_info, defrag);
C
Chris Mason 已提交
395 396 397 398 399 400 401 402 403 404
	}
	atomic_dec(&fs_info->defrag_running);

	/*
	 * during unmount, we use the transaction_wait queue to
	 * wait for the defragger to stop
	 */
	wake_up(&fs_info->transaction_wait);
	return 0;
}
C
Chris Mason 已提交
405

C
Chris Mason 已提交
406 407 408
/* simple helper to fault in pages and copy.  This should go away
 * and be replaced with calls into generic code.
 */
C
Chris Mason 已提交
409
static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
J
Josef Bacik 已提交
410
					 size_t write_bytes,
411
					 struct page **prepared_pages,
412
					 struct iov_iter *i)
C
Chris Mason 已提交
413
{
414
	size_t copied = 0;
J
Josef Bacik 已提交
415
	size_t total_copied = 0;
416
	int pg = 0;
C
Chris Mason 已提交
417 418
	int offset = pos & (PAGE_CACHE_SIZE - 1);

419
	while (write_bytes > 0) {
C
Chris Mason 已提交
420 421
		size_t count = min_t(size_t,
				     PAGE_CACHE_SIZE - offset, write_bytes);
422
		struct page *page = prepared_pages[pg];
423 424 425 426
		/*
		 * Copy data from userspace to the current page
		 */
		copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
427

C
Chris Mason 已提交
428 429
		/* Flush processor's dcache for this page */
		flush_dcache_page(page);
430 431 432 433 434 435 436 437 438 439 440 441 442

		/*
		 * if we get a partial write, we can end up with
		 * partially up to date pages.  These add
		 * a lot of complexity, so make sure they don't
		 * happen by forcing this copy to be retried.
		 *
		 * The rest of the btrfs_file_write code will fall
		 * back to page at a time copies after we return 0.
		 */
		if (!PageUptodate(page) && copied < count)
			copied = 0;

443 444
		iov_iter_advance(i, copied);
		write_bytes -= copied;
445
		total_copied += copied;
C
Chris Mason 已提交
446

A
Al Viro 已提交
447
		/* Return to btrfs_file_write_iter to fault page */
J
Josef Bacik 已提交
448
		if (unlikely(copied == 0))
449
			break;
450

451
		if (copied < PAGE_CACHE_SIZE - offset) {
452 453 454 455 456
			offset += copied;
		} else {
			pg++;
			offset = 0;
		}
C
Chris Mason 已提交
457
	}
458
	return total_copied;
C
Chris Mason 已提交
459 460
}

C
Chris Mason 已提交
461 462 463
/*
 * unlocks pages after btrfs_file_write is done with them
 */
464
static void btrfs_drop_pages(struct page **pages, size_t num_pages)
C
Chris Mason 已提交
465 466 467
{
	size_t i;
	for (i = 0; i < num_pages; i++) {
C
Chris Mason 已提交
468 469
		/* page checked is some magic around finding pages that
		 * have been modified without going through btrfs_set_page_dirty
470 471 472
		 * clear it here. There should be no need to mark the pages
		 * accessed as prepare_pages should have marked them accessed
		 * in prepare_pages via find_or_create_page()
C
Chris Mason 已提交
473
		 */
C
Chris Mason 已提交
474
		ClearPageChecked(pages[i]);
C
Chris Mason 已提交
475 476 477 478 479
		unlock_page(pages[i]);
		page_cache_release(pages[i]);
	}
}

C
Chris Mason 已提交
480 481 482 483 484 485 486 487
/*
 * after copy_from_user, pages need to be dirtied and we need to make
 * sure holes are created between the current EOF and the start of
 * any next extents (if required).
 *
 * this also makes the decision about creating an inline extent vs
 * doing real data extents, marking pages dirty and delalloc as required.
 */
488
int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
489 490 491
			     struct page **pages, size_t num_pages,
			     loff_t pos, size_t write_bytes,
			     struct extent_state **cached)
C
Chris Mason 已提交
492 493
{
	int err = 0;
494
	int i;
495
	u64 num_bytes;
496 497 498 499
	u64 start_pos;
	u64 end_of_last_block;
	u64 end_pos = pos + write_bytes;
	loff_t isize = i_size_read(inode);
C
Chris Mason 已提交
500

501
	start_pos = pos & ~((u64)root->sectorsize - 1);
502
	num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
C
Chris Mason 已提交
503

504
	end_of_last_block = start_pos + num_bytes - 1;
505
	err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
506
					cached);
J
Josef Bacik 已提交
507 508
	if (err)
		return err;
J
Josef Bacik 已提交
509

C
Chris Mason 已提交
510 511 512 513 514
	for (i = 0; i < num_pages; i++) {
		struct page *p = pages[i];
		SetPageUptodate(p);
		ClearPageChecked(p);
		set_page_dirty(p);
515
	}
J
Josef Bacik 已提交
516 517 518 519 520 521 522

	/*
	 * we've only changed i_size in ram, and we haven't updated
	 * the disk i_size.  There is no need to log the inode
	 * at this time.
	 */
	if (end_pos > isize)
523
		i_size_write(inode, end_pos);
524
	return 0;
C
Chris Mason 已提交
525 526
}

C
Chris Mason 已提交
527 528 529 530
/*
 * this drops all the extents in the cache that intersect the range
 * [start, end].  Existing extents are split as required.
 */
531 532
void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
			     int skip_pinned)
533 534
{
	struct extent_map *em;
535 536
	struct extent_map *split = NULL;
	struct extent_map *split2 = NULL;
537
	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
538
	u64 len = end - start + 1;
J
Josef Bacik 已提交
539
	u64 gen;
540 541
	int ret;
	int testend = 1;
542
	unsigned long flags;
C
Chris Mason 已提交
543
	int compressed = 0;
J
Josef Bacik 已提交
544
	bool modified;
545

546
	WARN_ON(end < start);
547
	if (end == (u64)-1) {
548
		len = (u64)-1;
549 550
		testend = 0;
	}
C
Chris Mason 已提交
551
	while (1) {
552 553
		int no_splits = 0;

J
Josef Bacik 已提交
554
		modified = false;
555
		if (!split)
556
			split = alloc_extent_map();
557
		if (!split2)
558
			split2 = alloc_extent_map();
559 560
		if (!split || !split2)
			no_splits = 1;
561

562
		write_lock(&em_tree->lock);
563
		em = lookup_extent_mapping(em_tree, start, len);
564
		if (!em) {
565
			write_unlock(&em_tree->lock);
566
			break;
567
		}
568
		flags = em->flags;
J
Josef Bacik 已提交
569
		gen = em->generation;
570
		if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
571
			if (testend && em->start + em->len >= start + len) {
572
				free_extent_map(em);
C
Chris Mason 已提交
573
				write_unlock(&em_tree->lock);
574 575
				break;
			}
576 577
			start = em->start + em->len;
			if (testend)
578 579
				len = start + len - (em->start + em->len);
			free_extent_map(em);
C
Chris Mason 已提交
580
			write_unlock(&em_tree->lock);
581 582
			continue;
		}
C
Chris Mason 已提交
583
		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
584
		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
L
Liu Bo 已提交
585
		clear_bit(EXTENT_FLAG_LOGGING, &flags);
J
Josef Bacik 已提交
586
		modified = !list_empty(&em->list);
587 588
		if (no_splits)
			goto next;
589

590
		if (em->start < start) {
591 592
			split->start = em->start;
			split->len = start - em->start;
593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612

			if (em->block_start < EXTENT_MAP_LAST_BYTE) {
				split->orig_start = em->orig_start;
				split->block_start = em->block_start;

				if (compressed)
					split->block_len = em->block_len;
				else
					split->block_len = split->len;
				split->orig_block_len = max(split->block_len,
						em->orig_block_len);
				split->ram_bytes = em->ram_bytes;
			} else {
				split->orig_start = split->start;
				split->block_len = 0;
				split->block_start = em->block_start;
				split->orig_block_len = 0;
				split->ram_bytes = split->len;
			}

J
Josef Bacik 已提交
613
			split->generation = gen;
614
			split->bdev = em->bdev;
615
			split->flags = flags;
616
			split->compress_type = em->compress_type;
617
			replace_extent_mapping(em_tree, em, split, modified);
618 619 620 621
			free_extent_map(split);
			split = split2;
			split2 = NULL;
		}
622
		if (testend && em->start + em->len > start + len) {
623 624 625 626 627
			u64 diff = start + len - em->start;

			split->start = start + len;
			split->len = em->start + em->len - (start + len);
			split->bdev = em->bdev;
628
			split->flags = flags;
629
			split->compress_type = em->compress_type;
J
Josef Bacik 已提交
630
			split->generation = gen;
631 632 633

			if (em->block_start < EXTENT_MAP_LAST_BYTE) {
				split->orig_block_len = max(em->block_len,
634
						    em->orig_block_len);
635

636 637 638 639 640 641 642 643 644 645 646
				split->ram_bytes = em->ram_bytes;
				if (compressed) {
					split->block_len = em->block_len;
					split->block_start = em->block_start;
					split->orig_start = em->orig_start;
				} else {
					split->block_len = split->len;
					split->block_start = em->block_start
						+ diff;
					split->orig_start = em->orig_start;
				}
C
Chris Mason 已提交
647
			} else {
648 649 650 651 652
				split->ram_bytes = split->len;
				split->orig_start = split->start;
				split->block_len = 0;
				split->block_start = em->block_start;
				split->orig_block_len = 0;
C
Chris Mason 已提交
653
			}
654

655 656 657 658 659 660 661 662
			if (extent_map_in_tree(em)) {
				replace_extent_mapping(em_tree, em, split,
						       modified);
			} else {
				ret = add_extent_mapping(em_tree, split,
							 modified);
				ASSERT(ret == 0); /* Logic error */
			}
663 664 665
			free_extent_map(split);
			split = NULL;
		}
666
next:
667 668
		if (extent_map_in_tree(em))
			remove_extent_mapping(em_tree, em);
669
		write_unlock(&em_tree->lock);
670

671 672 673 674 675
		/* once for us */
		free_extent_map(em);
		/* once for the tree*/
		free_extent_map(em);
	}
676 677 678 679
	if (split)
		free_extent_map(split);
	if (split2)
		free_extent_map(split2);
680 681
}

C
Chris Mason 已提交
682 683 684 685 686 687 688 689 690
/*
 * this is very complex, but the basic idea is to drop all extents
 * in the range start - end.  hint_block is filled in with a block number
 * that would be a good hint to the block allocator for this file.
 *
 * If an extent intersects the range but is not entirely inside the range
 * it is either truncated or split.  Anything entirely inside the range
 * is deleted from the tree.
 */
J
Josef Bacik 已提交
691 692 693
int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
			 struct btrfs_root *root, struct inode *inode,
			 struct btrfs_path *path, u64 start, u64 end,
694 695 696 697
			 u64 *drop_end, int drop_cache,
			 int replace_extent,
			 u32 extent_item_size,
			 int *key_inserted)
C
Chris Mason 已提交
698
{
699
	struct extent_buffer *leaf;
Y
Yan, Zheng 已提交
700
	struct btrfs_file_extent_item *fi;
701
	struct btrfs_key key;
Y
Yan, Zheng 已提交
702
	struct btrfs_key new_key;
L
Li Zefan 已提交
703
	u64 ino = btrfs_ino(inode);
Y
Yan, Zheng 已提交
704 705 706 707 708 709 710 711
	u64 search_start = start;
	u64 disk_bytenr = 0;
	u64 num_bytes = 0;
	u64 extent_offset = 0;
	u64 extent_end = 0;
	int del_nr = 0;
	int del_slot = 0;
	int extent_type;
C
Chris Mason 已提交
712
	int recow;
713
	int ret;
714
	int modify_tree = -1;
715
	int update_refs;
716
	int found = 0;
717
	int leafs_visited = 0;
C
Chris Mason 已提交
718

C
Chris Mason 已提交
719 720
	if (drop_cache)
		btrfs_drop_extent_cache(inode, start, end - 1, 0);
721

722
	if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
723 724
		modify_tree = 0;

725 726
	update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
		       root == root->fs_info->tree_root);
C
Chris Mason 已提交
727
	while (1) {
C
Chris Mason 已提交
728
		recow = 0;
L
Li Zefan 已提交
729
		ret = btrfs_lookup_file_extent(trans, root, path, ino,
730
					       search_start, modify_tree);
C
Chris Mason 已提交
731
		if (ret < 0)
Y
Yan, Zheng 已提交
732 733 734 735
			break;
		if (ret > 0 && path->slots[0] > 0 && search_start == start) {
			leaf = path->nodes[0];
			btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
L
Li Zefan 已提交
736
			if (key.objectid == ino &&
Y
Yan, Zheng 已提交
737 738
			    key.type == BTRFS_EXTENT_DATA_KEY)
				path->slots[0]--;
C
Chris Mason 已提交
739
		}
Y
Yan, Zheng 已提交
740
		ret = 0;
741
		leafs_visited++;
742
next_slot:
743
		leaf = path->nodes[0];
Y
Yan, Zheng 已提交
744 745 746 747 748 749 750 751
		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
			BUG_ON(del_nr > 0);
			ret = btrfs_next_leaf(root, path);
			if (ret < 0)
				break;
			if (ret > 0) {
				ret = 0;
				break;
752
			}
753
			leafs_visited++;
Y
Yan, Zheng 已提交
754 755 756 757 758
			leaf = path->nodes[0];
			recow = 1;
		}

		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
L
Li Zefan 已提交
759
		if (key.objectid > ino ||
Y
Yan, Zheng 已提交
760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775
		    key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
			break;

		fi = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_file_extent_item);
		extent_type = btrfs_file_extent_type(leaf, fi);

		if (extent_type == BTRFS_FILE_EXTENT_REG ||
		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
			num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
			extent_offset = btrfs_file_extent_offset(leaf, fi);
			extent_end = key.offset +
				btrfs_file_extent_num_bytes(leaf, fi);
		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
			extent_end = key.offset +
776 777
				btrfs_file_extent_inline_len(leaf,
						     path->slots[0], fi);
778
		} else {
Y
Yan, Zheng 已提交
779
			WARN_ON(1);
780
			extent_end = search_start;
C
Chris Mason 已提交
781 782
		}

783 784 785 786 787 788 789 790 791 792 793 794
		/*
		 * Don't skip extent items representing 0 byte lengths. They
		 * used to be created (bug) if while punching holes we hit
		 * -ENOSPC condition. So if we find one here, just ensure we
		 * delete it, otherwise we would insert a new file extent item
		 * with the same key (offset) as that 0 bytes length file
		 * extent item in the call to setup_items_for_insert() later
		 * in this function.
		 */
		if (extent_end == key.offset && extent_end >= search_start)
			goto delete_extent_item;

Y
Yan, Zheng 已提交
795 796
		if (extent_end <= search_start) {
			path->slots[0]++;
797
			goto next_slot;
C
Chris Mason 已提交
798 799
		}

800
		found = 1;
Y
Yan, Zheng 已提交
801
		search_start = max(key.offset, start);
802 803
		if (recow || !modify_tree) {
			modify_tree = -1;
804
			btrfs_release_path(path);
Y
Yan, Zheng 已提交
805
			continue;
C
Chris Mason 已提交
806
		}
Y
Yan Zheng 已提交
807

Y
Yan, Zheng 已提交
808 809 810 811 812 813
		/*
		 *     | - range to drop - |
		 *  | -------- extent -------- |
		 */
		if (start > key.offset && end < extent_end) {
			BUG_ON(del_nr > 0);
814
			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
815
				ret = -EOPNOTSUPP;
816 817
				break;
			}
Y
Yan, Zheng 已提交
818 819 820 821 822 823

			memcpy(&new_key, &key, sizeof(new_key));
			new_key.offset = start;
			ret = btrfs_duplicate_item(trans, root, path,
						   &new_key);
			if (ret == -EAGAIN) {
824
				btrfs_release_path(path);
Y
Yan, Zheng 已提交
825
				continue;
Y
Yan Zheng 已提交
826
			}
Y
Yan, Zheng 已提交
827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844
			if (ret < 0)
				break;

			leaf = path->nodes[0];
			fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
					    struct btrfs_file_extent_item);
			btrfs_set_file_extent_num_bytes(leaf, fi,
							start - key.offset);

			fi = btrfs_item_ptr(leaf, path->slots[0],
					    struct btrfs_file_extent_item);

			extent_offset += start - key.offset;
			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
			btrfs_set_file_extent_num_bytes(leaf, fi,
							extent_end - start);
			btrfs_mark_buffer_dirty(leaf);

J
Josef Bacik 已提交
845
			if (update_refs && disk_bytenr > 0) {
846
				ret = btrfs_inc_extent_ref(trans, root,
Y
Yan, Zheng 已提交
847 848 849
						disk_bytenr, num_bytes, 0,
						root->root_key.objectid,
						new_key.objectid,
J
Josef Bacik 已提交
850
						start - extent_offset, 1);
851
				BUG_ON(ret); /* -ENOMEM */
852
			}
Y
Yan, Zheng 已提交
853
			key.offset = start;
Y
Yan Zheng 已提交
854
		}
Y
Yan, Zheng 已提交
855 856 857 858 859
		/*
		 *  | ---- range to drop ----- |
		 *      | -------- extent -------- |
		 */
		if (start <= key.offset && end < extent_end) {
860
			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
861
				ret = -EOPNOTSUPP;
862 863
				break;
			}
Y
Yan Zheng 已提交
864

Y
Yan, Zheng 已提交
865 866
			memcpy(&new_key, &key, sizeof(new_key));
			new_key.offset = end;
867
			btrfs_set_item_key_safe(root->fs_info, path, &new_key);
Y
Yan Zheng 已提交
868

Y
Yan, Zheng 已提交
869 870 871 872 873
			extent_offset += end - key.offset;
			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
			btrfs_set_file_extent_num_bytes(leaf, fi,
							extent_end - end);
			btrfs_mark_buffer_dirty(leaf);
874
			if (update_refs && disk_bytenr > 0)
Y
Yan, Zheng 已提交
875 876
				inode_sub_bytes(inode, end - key.offset);
			break;
C
Chris Mason 已提交
877
		}
878

Y
Yan, Zheng 已提交
879 880 881 882 883 884 885
		search_start = extent_end;
		/*
		 *       | ---- range to drop ----- |
		 *  | -------- extent -------- |
		 */
		if (start > key.offset && end >= extent_end) {
			BUG_ON(del_nr > 0);
886
			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
887
				ret = -EOPNOTSUPP;
888 889
				break;
			}
890

Y
Yan, Zheng 已提交
891 892 893
			btrfs_set_file_extent_num_bytes(leaf, fi,
							start - key.offset);
			btrfs_mark_buffer_dirty(leaf);
894
			if (update_refs && disk_bytenr > 0)
Y
Yan, Zheng 已提交
895 896 897
				inode_sub_bytes(inode, extent_end - start);
			if (end == extent_end)
				break;
C
Chris Mason 已提交
898

Y
Yan, Zheng 已提交
899 900
			path->slots[0]++;
			goto next_slot;
Z
Zheng Yan 已提交
901 902
		}

Y
Yan, Zheng 已提交
903 904 905 906 907
		/*
		 *  | ---- range to drop ----- |
		 *    | ------ extent ------ |
		 */
		if (start <= key.offset && end >= extent_end) {
908
delete_extent_item:
Y
Yan, Zheng 已提交
909 910 911 912 913 914 915
			if (del_nr == 0) {
				del_slot = path->slots[0];
				del_nr = 1;
			} else {
				BUG_ON(del_slot + del_nr != path->slots[0]);
				del_nr++;
			}
Z
Zheng Yan 已提交
916

J
Josef Bacik 已提交
917 918
			if (update_refs &&
			    extent_type == BTRFS_FILE_EXTENT_INLINE) {
919
				inode_sub_bytes(inode,
Y
Yan, Zheng 已提交
920 921 922
						extent_end - key.offset);
				extent_end = ALIGN(extent_end,
						   root->sectorsize);
J
Josef Bacik 已提交
923
			} else if (update_refs && disk_bytenr > 0) {
Z
Zheng Yan 已提交
924
				ret = btrfs_free_extent(trans, root,
Y
Yan, Zheng 已提交
925 926
						disk_bytenr, num_bytes, 0,
						root->root_key.objectid,
927
						key.objectid, key.offset -
A
Arne Jansen 已提交
928
						extent_offset, 0);
929
				BUG_ON(ret); /* -ENOMEM */
Y
Yan, Zheng 已提交
930 931
				inode_sub_bytes(inode,
						extent_end - key.offset);
Z
Zheng Yan 已提交
932 933
			}

Y
Yan, Zheng 已提交
934 935 936 937 938 939 940 941 942 943
			if (end == extent_end)
				break;

			if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
				path->slots[0]++;
				goto next_slot;
			}

			ret = btrfs_del_items(trans, root, path, del_slot,
					      del_nr);
944 945
			if (ret) {
				btrfs_abort_transaction(trans, root, ret);
J
Josef Bacik 已提交
946
				break;
947
			}
Y
Yan, Zheng 已提交
948 949 950 951

			del_nr = 0;
			del_slot = 0;

952
			btrfs_release_path(path);
Y
Yan, Zheng 已提交
953
			continue;
C
Chris Mason 已提交
954
		}
Y
Yan, Zheng 已提交
955 956

		BUG_ON(1);
C
Chris Mason 已提交
957
	}
Y
Yan, Zheng 已提交
958

959
	if (!ret && del_nr > 0) {
960 961 962 963
		/*
		 * Set path->slots[0] to first slot, so that after the delete
		 * if items are move off from our leaf to its immediate left or
		 * right neighbor leafs, we end up with a correct and adjusted
964
		 * path->slots[0] for our insertion (if replace_extent != 0).
965 966
		 */
		path->slots[0] = del_slot;
Y
Yan, Zheng 已提交
967
		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
968 969
		if (ret)
			btrfs_abort_transaction(trans, root, ret);
970
	}
971

972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992
	leaf = path->nodes[0];
	/*
	 * If btrfs_del_items() was called, it might have deleted a leaf, in
	 * which case it unlocked our path, so check path->locks[0] matches a
	 * write lock.
	 */
	if (!ret && replace_extent && leafs_visited == 1 &&
	    (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
	     path->locks[0] == BTRFS_WRITE_LOCK) &&
	    btrfs_leaf_free_space(root, leaf) >=
	    sizeof(struct btrfs_item) + extent_item_size) {

		key.objectid = ino;
		key.type = BTRFS_EXTENT_DATA_KEY;
		key.offset = start;
		if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
			struct btrfs_key slot_key;

			btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
			if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
				path->slots[0]++;
993
		}
994 995 996 997 998 999
		setup_items_for_insert(root, path, &key,
				       &extent_item_size,
				       extent_item_size,
				       sizeof(struct btrfs_item) +
				       extent_item_size, 1);
		*key_inserted = 1;
Y
Yan Zheng 已提交
1000
	}
Y
Yan, Zheng 已提交
1001

1002 1003
	if (!replace_extent || !(*key_inserted))
		btrfs_release_path(path);
J
Josef Bacik 已提交
1004
	if (drop_end)
1005
		*drop_end = found ? min(end, extent_end) : end;
J
Josef Bacik 已提交
1006 1007 1008 1009 1010
	return ret;
}

int btrfs_drop_extents(struct btrfs_trans_handle *trans,
		       struct btrfs_root *root, struct inode *inode, u64 start,
1011
		       u64 end, int drop_cache)
J
Josef Bacik 已提交
1012 1013 1014 1015 1016 1017 1018
{
	struct btrfs_path *path;
	int ret;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;
J
Josef Bacik 已提交
1019
	ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1020
				   drop_cache, 0, 0, NULL);
Y
Yan, Zheng 已提交
1021
	btrfs_free_path(path);
C
Chris Mason 已提交
1022 1023 1024
	return ret;
}

Y
Yan Zheng 已提交
1025
static int extent_mergeable(struct extent_buffer *leaf, int slot,
1026 1027
			    u64 objectid, u64 bytenr, u64 orig_offset,
			    u64 *start, u64 *end)
Y
Yan Zheng 已提交
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042
{
	struct btrfs_file_extent_item *fi;
	struct btrfs_key key;
	u64 extent_end;

	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
		return 0;

	btrfs_item_key_to_cpu(leaf, &key, slot);
	if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
		return 0;

	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
	    btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1043
	    btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
Y
Yan Zheng 已提交
1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067
	    btrfs_file_extent_compression(leaf, fi) ||
	    btrfs_file_extent_encryption(leaf, fi) ||
	    btrfs_file_extent_other_encoding(leaf, fi))
		return 0;

	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
	if ((*start && *start != key.offset) || (*end && *end != extent_end))
		return 0;

	*start = key.offset;
	*end = extent_end;
	return 1;
}

/*
 * Mark extent in the range start - end as written.
 *
 * This changes extent type from 'pre-allocated' to 'regular'. If only
 * part of extent is marked as written, the extent will be split into
 * two or three.
 */
int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
			      struct inode *inode, u64 start, u64 end)
{
Y
Yan, Zheng 已提交
1068
	struct btrfs_root *root = BTRFS_I(inode)->root;
Y
Yan Zheng 已提交
1069 1070 1071 1072
	struct extent_buffer *leaf;
	struct btrfs_path *path;
	struct btrfs_file_extent_item *fi;
	struct btrfs_key key;
Y
Yan, Zheng 已提交
1073
	struct btrfs_key new_key;
Y
Yan Zheng 已提交
1074 1075 1076
	u64 bytenr;
	u64 num_bytes;
	u64 extent_end;
1077
	u64 orig_offset;
Y
Yan Zheng 已提交
1078 1079
	u64 other_start;
	u64 other_end;
Y
Yan, Zheng 已提交
1080 1081 1082
	u64 split;
	int del_nr = 0;
	int del_slot = 0;
1083
	int recow;
Y
Yan Zheng 已提交
1084
	int ret;
L
Li Zefan 已提交
1085
	u64 ino = btrfs_ino(inode);
Y
Yan Zheng 已提交
1086 1087

	path = btrfs_alloc_path();
1088 1089
	if (!path)
		return -ENOMEM;
Y
Yan Zheng 已提交
1090
again:
1091
	recow = 0;
Y
Yan, Zheng 已提交
1092
	split = start;
L
Li Zefan 已提交
1093
	key.objectid = ino;
Y
Yan Zheng 已提交
1094
	key.type = BTRFS_EXTENT_DATA_KEY;
Y
Yan, Zheng 已提交
1095
	key.offset = split;
Y
Yan Zheng 已提交
1096 1097

	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1098 1099
	if (ret < 0)
		goto out;
Y
Yan Zheng 已提交
1100 1101 1102 1103 1104
	if (ret > 0 && path->slots[0] > 0)
		path->slots[0]--;

	leaf = path->nodes[0];
	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
L
Li Zefan 已提交
1105
	BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
Y
Yan Zheng 已提交
1106 1107
	fi = btrfs_item_ptr(leaf, path->slots[0],
			    struct btrfs_file_extent_item);
Y
Yan, Zheng 已提交
1108 1109
	BUG_ON(btrfs_file_extent_type(leaf, fi) !=
	       BTRFS_FILE_EXTENT_PREALLOC);
Y
Yan Zheng 已提交
1110 1111 1112 1113 1114
	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
	BUG_ON(key.offset > start || extent_end < end);

	bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
	num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1115
	orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1116 1117 1118 1119 1120 1121
	memcpy(&new_key, &key, sizeof(new_key));

	if (start == key.offset && end < extent_end) {
		other_start = 0;
		other_end = start;
		if (extent_mergeable(leaf, path->slots[0] - 1,
L
Li Zefan 已提交
1122
				     ino, bytenr, orig_offset,
1123 1124
				     &other_start, &other_end)) {
			new_key.offset = end;
1125
			btrfs_set_item_key_safe(root->fs_info, path, &new_key);
1126 1127
			fi = btrfs_item_ptr(leaf, path->slots[0],
					    struct btrfs_file_extent_item);
1128 1129
			btrfs_set_file_extent_generation(leaf, fi,
							 trans->transid);
1130 1131 1132 1133 1134 1135
			btrfs_set_file_extent_num_bytes(leaf, fi,
							extent_end - end);
			btrfs_set_file_extent_offset(leaf, fi,
						     end - orig_offset);
			fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
					    struct btrfs_file_extent_item);
1136 1137
			btrfs_set_file_extent_generation(leaf, fi,
							 trans->transid);
1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148
			btrfs_set_file_extent_num_bytes(leaf, fi,
							end - other_start);
			btrfs_mark_buffer_dirty(leaf);
			goto out;
		}
	}

	if (start > key.offset && end == extent_end) {
		other_start = end;
		other_end = 0;
		if (extent_mergeable(leaf, path->slots[0] + 1,
L
Li Zefan 已提交
1149
				     ino, bytenr, orig_offset,
1150 1151 1152 1153 1154
				     &other_start, &other_end)) {
			fi = btrfs_item_ptr(leaf, path->slots[0],
					    struct btrfs_file_extent_item);
			btrfs_set_file_extent_num_bytes(leaf, fi,
							start - key.offset);
1155 1156
			btrfs_set_file_extent_generation(leaf, fi,
							 trans->transid);
1157 1158
			path->slots[0]++;
			new_key.offset = start;
1159
			btrfs_set_item_key_safe(root->fs_info, path, &new_key);
1160 1161 1162

			fi = btrfs_item_ptr(leaf, path->slots[0],
					    struct btrfs_file_extent_item);
1163 1164
			btrfs_set_file_extent_generation(leaf, fi,
							 trans->transid);
1165 1166 1167 1168 1169 1170 1171 1172
			btrfs_set_file_extent_num_bytes(leaf, fi,
							other_end - start);
			btrfs_set_file_extent_offset(leaf, fi,
						     start - orig_offset);
			btrfs_mark_buffer_dirty(leaf);
			goto out;
		}
	}
Y
Yan Zheng 已提交
1173

Y
Yan, Zheng 已提交
1174 1175 1176 1177 1178 1179 1180
	while (start > key.offset || end < extent_end) {
		if (key.offset == start)
			split = end;

		new_key.offset = split;
		ret = btrfs_duplicate_item(trans, root, path, &new_key);
		if (ret == -EAGAIN) {
1181
			btrfs_release_path(path);
Y
Yan, Zheng 已提交
1182
			goto again;
Y
Yan Zheng 已提交
1183
		}
1184 1185 1186 1187
		if (ret < 0) {
			btrfs_abort_transaction(trans, root, ret);
			goto out;
		}
Y
Yan Zheng 已提交
1188

Y
Yan, Zheng 已提交
1189 1190
		leaf = path->nodes[0];
		fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
Y
Yan Zheng 已提交
1191
				    struct btrfs_file_extent_item);
1192
		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
Y
Yan Zheng 已提交
1193
		btrfs_set_file_extent_num_bytes(leaf, fi,
Y
Yan, Zheng 已提交
1194 1195 1196 1197 1198
						split - key.offset);

		fi = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_file_extent_item);

1199
		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
Y
Yan, Zheng 已提交
1200 1201 1202
		btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
		btrfs_set_file_extent_num_bytes(leaf, fi,
						extent_end - split);
Y
Yan Zheng 已提交
1203 1204
		btrfs_mark_buffer_dirty(leaf);

Y
Yan, Zheng 已提交
1205 1206
		ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
					   root->root_key.objectid,
J
Josef Bacik 已提交
1207
					   ino, orig_offset, 1);
1208
		BUG_ON(ret); /* -ENOMEM */
Y
Yan Zheng 已提交
1209

Y
Yan, Zheng 已提交
1210 1211 1212 1213
		if (split == start) {
			key.offset = start;
		} else {
			BUG_ON(start != key.offset);
Y
Yan Zheng 已提交
1214
			path->slots[0]--;
Y
Yan, Zheng 已提交
1215
			extent_end = end;
Y
Yan Zheng 已提交
1216
		}
1217
		recow = 1;
Y
Yan Zheng 已提交
1218 1219
	}

Y
Yan, Zheng 已提交
1220 1221
	other_start = end;
	other_end = 0;
1222
	if (extent_mergeable(leaf, path->slots[0] + 1,
L
Li Zefan 已提交
1223
			     ino, bytenr, orig_offset,
1224 1225
			     &other_start, &other_end)) {
		if (recow) {
1226
			btrfs_release_path(path);
1227 1228
			goto again;
		}
Y
Yan, Zheng 已提交
1229 1230 1231 1232 1233
		extent_end = other_end;
		del_slot = path->slots[0] + 1;
		del_nr++;
		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
					0, root->root_key.objectid,
A
Arne Jansen 已提交
1234
					ino, orig_offset, 0);
1235
		BUG_ON(ret); /* -ENOMEM */
Y
Yan Zheng 已提交
1236
	}
Y
Yan, Zheng 已提交
1237 1238
	other_start = 0;
	other_end = start;
1239
	if (extent_mergeable(leaf, path->slots[0] - 1,
L
Li Zefan 已提交
1240
			     ino, bytenr, orig_offset,
1241 1242
			     &other_start, &other_end)) {
		if (recow) {
1243
			btrfs_release_path(path);
1244 1245
			goto again;
		}
Y
Yan, Zheng 已提交
1246 1247 1248 1249 1250
		key.offset = other_start;
		del_slot = path->slots[0];
		del_nr++;
		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
					0, root->root_key.objectid,
A
Arne Jansen 已提交
1251
					ino, orig_offset, 0);
1252
		BUG_ON(ret); /* -ENOMEM */
Y
Yan, Zheng 已提交
1253 1254
	}
	if (del_nr == 0) {
1255 1256
		fi = btrfs_item_ptr(leaf, path->slots[0],
			   struct btrfs_file_extent_item);
Y
Yan, Zheng 已提交
1257 1258
		btrfs_set_file_extent_type(leaf, fi,
					   BTRFS_FILE_EXTENT_REG);
1259
		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
Y
Yan, Zheng 已提交
1260
		btrfs_mark_buffer_dirty(leaf);
1261
	} else {
1262 1263
		fi = btrfs_item_ptr(leaf, del_slot - 1,
			   struct btrfs_file_extent_item);
1264 1265
		btrfs_set_file_extent_type(leaf, fi,
					   BTRFS_FILE_EXTENT_REG);
1266
		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1267 1268 1269
		btrfs_set_file_extent_num_bytes(leaf, fi,
						extent_end - key.offset);
		btrfs_mark_buffer_dirty(leaf);
Y
Yan, Zheng 已提交
1270

1271
		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1272 1273 1274 1275
		if (ret < 0) {
			btrfs_abort_transaction(trans, root, ret);
			goto out;
		}
1276
	}
Y
Yan, Zheng 已提交
1277
out:
Y
Yan Zheng 已提交
1278 1279 1280 1281
	btrfs_free_path(path);
	return 0;
}

1282 1283 1284 1285
/*
 * on error we return an unlocked page and the error value
 * on success we return a locked page and 0
 */
1286 1287
static int prepare_uptodate_page(struct page *page, u64 pos,
				 bool force_uptodate)
1288 1289 1290
{
	int ret = 0;

1291 1292
	if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
	    !PageUptodate(page)) {
1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304
		ret = btrfs_readpage(NULL, page);
		if (ret)
			return ret;
		lock_page(page);
		if (!PageUptodate(page)) {
			unlock_page(page);
			return -EIO;
		}
	}
	return 0;
}

C
Chris Mason 已提交
1305
/*
1306
 * this just gets pages into the page cache and locks them down.
C
Chris Mason 已提交
1307
 */
1308 1309 1310
static noinline int prepare_pages(struct inode *inode, struct page **pages,
				  size_t num_pages, loff_t pos,
				  size_t write_bytes, bool force_uptodate)
C
Chris Mason 已提交
1311 1312 1313
{
	int i;
	unsigned long index = pos >> PAGE_CACHE_SHIFT;
1314
	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1315
	int err = 0;
1316
	int faili;
1317

C
Chris Mason 已提交
1318
	for (i = 0; i < num_pages; i++) {
1319
		pages[i] = find_or_create_page(inode->i_mapping, index + i,
1320
					       mask | __GFP_WRITE);
C
Chris Mason 已提交
1321
		if (!pages[i]) {
1322 1323 1324 1325 1326 1327
			faili = i - 1;
			err = -ENOMEM;
			goto fail;
		}

		if (i == 0)
1328 1329
			err = prepare_uptodate_page(pages[i], pos,
						    force_uptodate);
1330 1331
		if (i == num_pages - 1)
			err = prepare_uptodate_page(pages[i],
1332
						    pos + write_bytes, false);
1333 1334 1335 1336
		if (err) {
			page_cache_release(pages[i]);
			faili = i - 1;
			goto fail;
C
Chris Mason 已提交
1337
		}
C
Chris Mason 已提交
1338
		wait_on_page_writeback(pages[i]);
C
Chris Mason 已提交
1339
	}
1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375

	return 0;
fail:
	while (faili >= 0) {
		unlock_page(pages[faili]);
		page_cache_release(pages[faili]);
		faili--;
	}
	return err;

}

/*
 * This function locks the extent and properly waits for data=ordered extents
 * to finish before allowing the pages to be modified if need.
 *
 * The return value:
 * 1 - the extent is locked
 * 0 - the extent is not locked, and everything is OK
 * -EAGAIN - need re-prepare the pages
 * the other < 0 number - Something wrong happens
 */
static noinline int
lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
				size_t num_pages, loff_t pos,
				u64 *lockstart, u64 *lockend,
				struct extent_state **cached_state)
{
	u64 start_pos;
	u64 last_pos;
	int i;
	int ret = 0;

	start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
	last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;

1376
	if (start_pos < inode->i_size) {
1377
		struct btrfs_ordered_extent *ordered;
1378
		lock_extent_bits(&BTRFS_I(inode)->io_tree,
1379
				 start_pos, last_pos, 0, cached_state);
1380 1381
		ordered = btrfs_lookup_ordered_range(inode, start_pos,
						     last_pos - start_pos + 1);
1382 1383
		if (ordered &&
		    ordered->file_offset + ordered->len > start_pos &&
1384
		    ordered->file_offset <= last_pos) {
1385
			unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1386 1387
					     start_pos, last_pos,
					     cached_state, GFP_NOFS);
1388 1389 1390 1391
			for (i = 0; i < num_pages; i++) {
				unlock_page(pages[i]);
				page_cache_release(pages[i]);
			}
1392 1393 1394
			btrfs_start_ordered_extent(inode, ordered, 1);
			btrfs_put_ordered_extent(ordered);
			return -EAGAIN;
1395 1396 1397 1398
		}
		if (ordered)
			btrfs_put_ordered_extent(ordered);

1399
		clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1400
				  last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1401
				  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1402 1403 1404 1405
				  0, 0, cached_state, GFP_NOFS);
		*lockstart = start_pos;
		*lockend = last_pos;
		ret = 1;
1406
	}
1407

1408
	for (i = 0; i < num_pages; i++) {
1409 1410
		if (clear_page_dirty_for_io(pages[i]))
			account_page_redirty(pages[i]);
1411 1412 1413
		set_page_extent_mapped(pages[i]);
		WARN_ON(!PageLocked(pages[i]));
	}
1414

1415
	return ret;
C
Chris Mason 已提交
1416 1417
}

1418 1419 1420 1421 1422 1423 1424 1425 1426
static noinline int check_can_nocow(struct inode *inode, loff_t pos,
				    size_t *write_bytes)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_ordered_extent *ordered;
	u64 lockstart, lockend;
	u64 num_bytes;
	int ret;

1427
	ret = btrfs_start_write_no_snapshoting(root);
1428 1429 1430
	if (!ret)
		return -ENOSPC;

1431
	lockstart = round_down(pos, root->sectorsize);
1432
	lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446

	while (1) {
		lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
		ordered = btrfs_lookup_ordered_range(inode, lockstart,
						     lockend - lockstart + 1);
		if (!ordered) {
			break;
		}
		unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
		btrfs_start_ordered_extent(inode, ordered, 1);
		btrfs_put_ordered_extent(ordered);
	}

	num_bytes = lockend - lockstart + 1;
1447
	ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1448 1449
	if (ret <= 0) {
		ret = 0;
1450
		btrfs_end_write_no_snapshoting(root);
1451
	} else {
1452 1453
		*write_bytes = min_t(size_t, *write_bytes ,
				     num_bytes - pos + lockstart);
1454 1455 1456 1457 1458 1459 1460
	}

	unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);

	return ret;
}

J
Josef Bacik 已提交
1461 1462 1463
static noinline ssize_t __btrfs_buffered_write(struct file *file,
					       struct iov_iter *i,
					       loff_t pos)
1464
{
A
Al Viro 已提交
1465
	struct inode *inode = file_inode(file);
1466 1467
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct page **pages = NULL;
1468
	struct extent_state *cached_state = NULL;
1469
	u64 release_bytes = 0;
1470 1471
	u64 lockstart;
	u64 lockend;
C
Chris Mason 已提交
1472
	unsigned long first_index;
J
Josef Bacik 已提交
1473 1474
	size_t num_written = 0;
	int nrptrs;
1475
	int ret = 0;
1476
	bool only_release_metadata = false;
1477
	bool force_page_uptodate = false;
1478
	bool need_unlock;
1479

1480 1481
	nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_CACHE_SIZE),
			PAGE_CACHE_SIZE / (sizeof(struct page *)));
1482 1483
	nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
	nrptrs = max(nrptrs, 8);
1484
	pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
J
Josef Bacik 已提交
1485 1486
	if (!pages)
		return -ENOMEM;
1487

C
Chris Mason 已提交
1488 1489
	first_index = pos >> PAGE_CACHE_SHIFT;

J
Josef Bacik 已提交
1490
	while (iov_iter_count(i) > 0) {
C
Chris Mason 已提交
1491
		size_t offset = pos & (PAGE_CACHE_SIZE - 1);
J
Josef Bacik 已提交
1492
		size_t write_bytes = min(iov_iter_count(i),
1493
					 nrptrs * (size_t)PAGE_CACHE_SIZE -
1494
					 offset);
1495 1496
		size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
						PAGE_CACHE_SIZE);
1497
		size_t reserve_bytes;
J
Josef Bacik 已提交
1498 1499
		size_t dirty_pages;
		size_t copied;
C
Chris Mason 已提交
1500

1501
		WARN_ON(num_pages > nrptrs);
1502

1503 1504 1505 1506
		/*
		 * Fault pages before locking them in prepare_pages
		 * to avoid recursive lock
		 */
J
Josef Bacik 已提交
1507
		if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1508
			ret = -EFAULT;
J
Josef Bacik 已提交
1509
			break;
1510 1511
		}

1512
		reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1513
		ret = btrfs_check_data_free_space(inode, reserve_bytes, write_bytes);
1514 1515 1516 1517 1518 1519 1520 1521 1522 1523
		if (ret == -ENOSPC &&
		    (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
					      BTRFS_INODE_PREALLOC))) {
			ret = check_can_nocow(inode, pos, &write_bytes);
			if (ret > 0) {
				only_release_metadata = true;
				/*
				 * our prealloc extent may be smaller than
				 * write_bytes, so scale down.
				 */
1524 1525
				num_pages = DIV_ROUND_UP(write_bytes + offset,
							 PAGE_CACHE_SIZE);
1526 1527 1528 1529 1530 1531 1532
				reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
				ret = 0;
			} else {
				ret = -ENOSPC;
			}
		}

1533
		if (ret)
J
Josef Bacik 已提交
1534
			break;
1535

1536 1537 1538 1539 1540
		ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
		if (ret) {
			if (!only_release_metadata)
				btrfs_free_reserved_data_space(inode,
							       reserve_bytes);
1541
			else
1542
				btrfs_end_write_no_snapshoting(root);
1543 1544 1545 1546
			break;
		}

		release_bytes = reserve_bytes;
1547 1548
		need_unlock = false;
again:
1549 1550 1551 1552 1553
		/*
		 * This is going to setup the pages array with the number of
		 * pages we want, so we don't really need to worry about the
		 * contents of pages from loop to loop
		 */
1554 1555
		ret = prepare_pages(inode, pages, num_pages,
				    pos, write_bytes,
1556
				    force_page_uptodate);
1557
		if (ret)
J
Josef Bacik 已提交
1558
			break;
C
Chris Mason 已提交
1559

1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571
		ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
						      pos, &lockstart, &lockend,
						      &cached_state);
		if (ret < 0) {
			if (ret == -EAGAIN)
				goto again;
			break;
		} else if (ret > 0) {
			need_unlock = true;
			ret = 0;
		}

1572
		copied = btrfs_copy_from_user(pos, num_pages,
J
Josef Bacik 已提交
1573
					   write_bytes, pages, i);
1574 1575 1576 1577 1578 1579 1580 1581

		/*
		 * if we have trouble faulting in the pages, fall
		 * back to one page at a time
		 */
		if (copied < write_bytes)
			nrptrs = 1;

1582 1583
		if (copied == 0) {
			force_page_uptodate = true;
1584
			dirty_pages = 0;
1585 1586
		} else {
			force_page_uptodate = false;
1587 1588
			dirty_pages = DIV_ROUND_UP(copied + offset,
						   PAGE_CACHE_SIZE);
1589
		}
1590

J
Josef Bacik 已提交
1591 1592 1593 1594 1595 1596 1597
		/*
		 * If we had a short copy we need to release the excess delaloc
		 * bytes we reserved.  We need to increment outstanding_extents
		 * because btrfs_delalloc_release_space will decrement it, but
		 * we still have an outstanding extent for the chunk we actually
		 * managed to copy.
		 */
1598
		if (num_pages > dirty_pages) {
1599 1600
			release_bytes = (num_pages - dirty_pages) <<
				PAGE_CACHE_SHIFT;
1601 1602 1603 1604 1605
			if (copied > 0) {
				spin_lock(&BTRFS_I(inode)->lock);
				BTRFS_I(inode)->outstanding_extents++;
				spin_unlock(&BTRFS_I(inode)->lock);
			}
1606 1607 1608 1609 1610 1611
			if (only_release_metadata)
				btrfs_delalloc_release_metadata(inode,
								release_bytes);
			else
				btrfs_delalloc_release_space(inode,
							     release_bytes);
1612 1613
		}

1614
		release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1615 1616

		if (copied > 0)
1617 1618 1619
			ret = btrfs_dirty_pages(root, inode, pages,
						dirty_pages, pos, copied,
						NULL);
1620 1621 1622 1623
		if (need_unlock)
			unlock_extent_cached(&BTRFS_I(inode)->io_tree,
					     lockstart, lockend, &cached_state,
					     GFP_NOFS);
1624 1625
		if (ret) {
			btrfs_drop_pages(pages, num_pages);
1626
			break;
1627
		}
C
Chris Mason 已提交
1628

1629
		release_bytes = 0;
1630
		if (only_release_metadata)
1631
			btrfs_end_write_no_snapshoting(root);
1632

1633
		if (only_release_metadata && copied > 0) {
1634 1635
			lockstart = round_down(pos, root->sectorsize);
			lockend = lockstart +
1636 1637 1638 1639 1640 1641 1642 1643
				(dirty_pages << PAGE_CACHE_SHIFT) - 1;

			set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
				       lockend, EXTENT_NORESERVE, NULL,
				       NULL, GFP_NOFS);
			only_release_metadata = false;
		}

1644 1645
		btrfs_drop_pages(pages, num_pages);

J
Josef Bacik 已提交
1646 1647
		cond_resched();

1648
		balance_dirty_pages_ratelimited(inode->i_mapping);
1649
		if (dirty_pages < (root->nodesize >> PAGE_CACHE_SHIFT) + 1)
1650
			btrfs_btree_balance_dirty(root);
1651

1652 1653
		pos += copied;
		num_written += copied;
J
Josef Bacik 已提交
1654
	}
C
Chris Mason 已提交
1655

J
Josef Bacik 已提交
1656 1657
	kfree(pages);

1658
	if (release_bytes) {
1659
		if (only_release_metadata) {
1660
			btrfs_end_write_no_snapshoting(root);
1661
			btrfs_delalloc_release_metadata(inode, release_bytes);
1662
		} else {
1663
			btrfs_delalloc_release_space(inode, release_bytes);
1664
		}
1665 1666
	}

J
Josef Bacik 已提交
1667 1668 1669 1670
	return num_written ? num_written : ret;
}

static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1671
				    struct iov_iter *from,
A
Al Viro 已提交
1672
				    loff_t pos)
J
Josef Bacik 已提交
1673 1674
{
	struct file *file = iocb->ki_filp;
1675
	struct inode *inode = file_inode(file);
J
Josef Bacik 已提交
1676 1677 1678 1679 1680
	ssize_t written;
	ssize_t written_buffered;
	loff_t endbyte;
	int err;

A
Al Viro 已提交
1681
	written = generic_file_direct_write(iocb, from, pos);
J
Josef Bacik 已提交
1682

A
Al Viro 已提交
1683
	if (written < 0 || !iov_iter_count(from))
J
Josef Bacik 已提交
1684 1685 1686
		return written;

	pos += written;
1687
	written_buffered = __btrfs_buffered_write(file, from, pos);
J
Josef Bacik 已提交
1688 1689 1690
	if (written_buffered < 0) {
		err = written_buffered;
		goto out;
C
Chris Mason 已提交
1691
	}
1692 1693 1694 1695
	/*
	 * Ensure all data is persisted. We want the next direct IO read to be
	 * able to read what was just written.
	 */
J
Josef Bacik 已提交
1696
	endbyte = pos + written_buffered - 1;
1697
	err = btrfs_fdatawrite_range(inode, pos, endbyte);
1698 1699
	if (err)
		goto out;
1700
	err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
J
Josef Bacik 已提交
1701 1702 1703
	if (err)
		goto out;
	written += written_buffered;
1704
	iocb->ki_pos = pos + written_buffered;
J
Josef Bacik 已提交
1705 1706
	invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
				 endbyte >> PAGE_CACHE_SHIFT);
C
Chris Mason 已提交
1707
out:
J
Josef Bacik 已提交
1708 1709
	return written ? written : err;
}
1710

1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728
static void update_time_for_write(struct inode *inode)
{
	struct timespec now;

	if (IS_NOCMTIME(inode))
		return;

	now = current_fs_time(inode->i_sb);
	if (!timespec_equal(&inode->i_mtime, &now))
		inode->i_mtime = now;

	if (!timespec_equal(&inode->i_ctime, &now))
		inode->i_ctime = now;

	if (IS_I_VERSION(inode))
		inode_inc_iversion(inode);
}

A
Al Viro 已提交
1729 1730
static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
				    struct iov_iter *from)
J
Josef Bacik 已提交
1731 1732
{
	struct file *file = iocb->ki_filp;
A
Al Viro 已提交
1733
	struct inode *inode = file_inode(file);
J
Josef Bacik 已提交
1734
	struct btrfs_root *root = BTRFS_I(inode)->root;
1735
	u64 start_pos;
1736
	u64 end_pos;
J
Josef Bacik 已提交
1737
	ssize_t num_written = 0;
1738
	bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1739 1740 1741
	ssize_t err;
	loff_t pos;
	size_t count;
J
Josef Bacik 已提交
1742 1743

	mutex_lock(&inode->i_mutex);
1744 1745
	err = generic_write_checks(iocb, from);
	if (err <= 0) {
J
Josef Bacik 已提交
1746
		mutex_unlock(&inode->i_mutex);
1747
		return err;
J
Josef Bacik 已提交
1748 1749
	}

1750
	current->backing_dev_info = inode_to_bdi(inode);
1751
	err = file_remove_privs(file);
J
Josef Bacik 已提交
1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762
	if (err) {
		mutex_unlock(&inode->i_mutex);
		goto out;
	}

	/*
	 * If BTRFS flips readonly due to some impossible error
	 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
	 * although we have opened a file as writable, we have
	 * to stop this write operation to ensure FS consistency.
	 */
1763
	if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
J
Josef Bacik 已提交
1764 1765 1766 1767 1768
		mutex_unlock(&inode->i_mutex);
		err = -EROFS;
		goto out;
	}

1769 1770 1771 1772 1773 1774 1775
	/*
	 * We reserve space for updating the inode when we reserve space for the
	 * extent we are going to write, so we will enospc out there.  We don't
	 * need to start yet another transaction to update the inode as we will
	 * update the inode when we finish writing whatever data we write.
	 */
	update_time_for_write(inode);
J
Josef Bacik 已提交
1776

1777 1778
	pos = iocb->ki_pos;
	count = iov_iter_count(from);
1779 1780
	start_pos = round_down(pos, root->sectorsize);
	if (start_pos > i_size_read(inode)) {
1781
		/* Expand hole size to cover write data, preventing empty gap */
1782
		end_pos = round_up(pos + count, root->sectorsize);
1783
		err = btrfs_cont_expand(inode, i_size_read(inode), end_pos);
1784 1785 1786 1787 1788 1789
		if (err) {
			mutex_unlock(&inode->i_mutex);
			goto out;
		}
	}

1790 1791 1792
	if (sync)
		atomic_inc(&BTRFS_I(inode)->sync_writers);

1793
	if (iocb->ki_flags & IOCB_DIRECT) {
A
Al Viro 已提交
1794
		num_written = __btrfs_direct_write(iocb, from, pos);
J
Josef Bacik 已提交
1795
	} else {
A
Al Viro 已提交
1796
		num_written = __btrfs_buffered_write(file, from, pos);
J
Josef Bacik 已提交
1797
		if (num_written > 0)
1798
			iocb->ki_pos = pos + num_written;
J
Josef Bacik 已提交
1799 1800 1801
	}

	mutex_unlock(&inode->i_mutex);
1802

1803
	/*
1804 1805 1806
	 * We also have to set last_sub_trans to the current log transid,
	 * otherwise subsequent syncs to a file that's been synced in this
	 * transaction will appear to have already occured.
1807
	 */
1808
	spin_lock(&BTRFS_I(inode)->lock);
1809
	BTRFS_I(inode)->last_sub_trans = root->log_transid;
1810
	spin_unlock(&BTRFS_I(inode)->lock);
1811
	if (num_written > 0) {
J
Josef Bacik 已提交
1812
		err = generic_write_sync(file, pos, num_written);
1813
		if (err < 0)
1814 1815
			num_written = err;
	}
1816

1817 1818
	if (sync)
		atomic_dec(&BTRFS_I(inode)->sync_writers);
1819
out:
C
Chris Mason 已提交
1820 1821 1822 1823
	current->backing_dev_info = NULL;
	return num_written ? num_written : err;
}

C
Chris Mason 已提交
1824
int btrfs_release_file(struct inode *inode, struct file *filp)
1825
{
S
Sage Weil 已提交
1826 1827
	if (filp->private_data)
		btrfs_ioctl_trans_end(filp);
1828 1829 1830 1831 1832 1833 1834 1835 1836
	/*
	 * ordered_data_close is set by settattr when we are about to truncate
	 * a file from a non-zero size to a zero size.  This tries to
	 * flush down new bytes that may have been written if the
	 * application were using truncate to replace a file in place.
	 */
	if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
			       &BTRFS_I(inode)->runtime_flags))
			filemap_flush(inode->i_mapping);
1837 1838 1839
	return 0;
}

1840 1841 1842 1843 1844
static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
{
	int ret;

	atomic_inc(&BTRFS_I(inode)->sync_writers);
1845
	ret = btrfs_fdatawrite_range(inode, start, end);
1846 1847 1848 1849 1850
	atomic_dec(&BTRFS_I(inode)->sync_writers);

	return ret;
}

C
Chris Mason 已提交
1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861
/*
 * fsync call for both files and directories.  This logs the inode into
 * the tree log instead of forcing full commits whenever possible.
 *
 * It needs to call filemap_fdatawait so that all ordered extent updates are
 * in the metadata btree are up to date for copying to the log.
 *
 * It drops the inode mutex before doing the tree log commit.  This is an
 * important optimization for directories because holding the mutex prevents
 * new operations on the dir while we write to disk.
 */
1862
int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
C
Chris Mason 已提交
1863
{
1864
	struct dentry *dentry = file->f_path.dentry;
1865
	struct inode *inode = d_inode(dentry);
C
Chris Mason 已提交
1866 1867
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_trans_handle *trans;
1868 1869
	struct btrfs_log_ctx ctx;
	int ret = 0;
1870
	bool full_sync = 0;
1871
	const u64 len = end - start + 1;
C
Chris Mason 已提交
1872

1873
	trace_btrfs_sync_file(file, datasync);
1874

1875 1876 1877
	/*
	 * We write the dirty pages in the range and wait until they complete
	 * out of the ->i_mutex. If so, we can flush the dirty pages by
1878 1879
	 * multi-task, and make the performance up.  See
	 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1880
	 */
1881
	ret = start_ordered_ops(inode, start, end);
1882 1883 1884
	if (ret)
		return ret;

1885
	mutex_lock(&inode->i_mutex);
M
Miao Xie 已提交
1886
	atomic_inc(&root->log_batch);
1887 1888
	full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
			     &BTRFS_I(inode)->runtime_flags);
1889 1890 1891 1892
	/*
	 * We might have have had more pages made dirty after calling
	 * start_ordered_ops and before acquiring the inode's i_mutex.
	 */
1893
	if (full_sync) {
1894 1895 1896 1897 1898 1899
		/*
		 * For a full sync, we need to make sure any ordered operations
		 * start and finish before we start logging the inode, so that
		 * all extents are persisted and the respective file extent
		 * items are in the fs/subvol btree.
		 */
1900
		ret = btrfs_wait_ordered_range(inode, start, len);
1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938
	} else {
		/*
		 * Start any new ordered operations before starting to log the
		 * inode. We will wait for them to finish in btrfs_sync_log().
		 *
		 * Right before acquiring the inode's mutex, we might have new
		 * writes dirtying pages, which won't immediately start the
		 * respective ordered operations - that is done through the
		 * fill_delalloc callbacks invoked from the writepage and
		 * writepages address space operations. So make sure we start
		 * all ordered operations before starting to log our inode. Not
		 * doing this means that while logging the inode, writeback
		 * could start and invoke writepage/writepages, which would call
		 * the fill_delalloc callbacks (cow_file_range,
		 * submit_compressed_extents). These callbacks add first an
		 * extent map to the modified list of extents and then create
		 * the respective ordered operation, which means in
		 * tree-log.c:btrfs_log_inode() we might capture all existing
		 * ordered operations (with btrfs_get_logged_extents()) before
		 * the fill_delalloc callback adds its ordered operation, and by
		 * the time we visit the modified list of extent maps (with
		 * btrfs_log_changed_extents()), we see and process the extent
		 * map they created. We then use the extent map to construct a
		 * file extent item for logging without waiting for the
		 * respective ordered operation to finish - this file extent
		 * item points to a disk location that might not have yet been
		 * written to, containing random data - so after a crash a log
		 * replay will make our inode have file extent items that point
		 * to disk locations containing invalid data, as we returned
		 * success to userspace without waiting for the respective
		 * ordered operation to finish, because it wasn't captured by
		 * btrfs_get_logged_extents().
		 */
		ret = start_ordered_ops(inode, start, end);
	}
	if (ret) {
		mutex_unlock(&inode->i_mutex);
		goto out;
1939
	}
M
Miao Xie 已提交
1940
	atomic_inc(&root->log_batch);
1941

C
Chris Mason 已提交
1942
	/*
1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968
	 * If the last transaction that changed this file was before the current
	 * transaction and we have the full sync flag set in our inode, we can
	 * bail out now without any syncing.
	 *
	 * Note that we can't bail out if the full sync flag isn't set. This is
	 * because when the full sync flag is set we start all ordered extents
	 * and wait for them to fully complete - when they complete they update
	 * the inode's last_trans field through:
	 *
	 *     btrfs_finish_ordered_io() ->
	 *         btrfs_update_inode_fallback() ->
	 *             btrfs_update_inode() ->
	 *                 btrfs_set_inode_last_trans()
	 *
	 * So we are sure that last_trans is up to date and can do this check to
	 * bail out safely. For the fast path, when the full sync flag is not
	 * set in our inode, we can not do it because we start only our ordered
	 * extents and don't wait for them to complete (that is when
	 * btrfs_finish_ordered_io runs), so here at this point their last_trans
	 * value might be less than or equals to fs_info->last_trans_committed,
	 * and setting a speculative last_trans for an inode when a buffered
	 * write is made (such as fs_info->generation + 1 for example) would not
	 * be reliable since after setting the value and before fsync is called
	 * any number of transactions can start and commit (transaction kthread
	 * commits the current transaction periodically), and a transaction
	 * commit does not start nor waits for ordered extents to complete.
1969
	 */
J
Josef Bacik 已提交
1970
	smp_mb();
1971
	if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1972 1973 1974 1975
	    (BTRFS_I(inode)->last_trans <=
	     root->fs_info->last_trans_committed &&
	     (full_sync ||
	      !btrfs_have_ordered_extents_in_range(inode, start, len)))) {
J
Josef Bacik 已提交
1976 1977 1978 1979 1980 1981 1982
		/*
		 * We'v had everything committed since the last time we were
		 * modified so clear this flag in case it was set for whatever
		 * reason, it's no longer relevant.
		 */
		clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
			  &BTRFS_I(inode)->runtime_flags);
1983
		mutex_unlock(&inode->i_mutex);
1984 1985 1986 1987
		goto out;
	}

	/*
1988 1989
	 * ok we haven't committed the transaction yet, lets do a commit
	 */
1990
	if (file->private_data)
S
Sage Weil 已提交
1991 1992
		btrfs_ioctl_trans_end(file);

1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
	/*
	 * We use start here because we will need to wait on the IO to complete
	 * in btrfs_sync_log, which could require joining a transaction (for
	 * example checking cross references in the nocow path).  If we use join
	 * here we could get into a situation where we're waiting on IO to
	 * happen that is blocked on a transaction trying to commit.  With start
	 * we inc the extwriter counter, so we wait for all extwriters to exit
	 * before we start blocking join'ers.  This comment is to keep somebody
	 * from thinking they are super smart and changing this to
	 * btrfs_join_transaction *cough*Josef*cough*.
	 */
2004 2005 2006
	trans = btrfs_start_transaction(root, 0);
	if (IS_ERR(trans)) {
		ret = PTR_ERR(trans);
2007
		mutex_unlock(&inode->i_mutex);
C
Chris Mason 已提交
2008 2009
		goto out;
	}
2010
	trans->sync = true;
2011

2012 2013
	btrfs_init_log_ctx(&ctx);

2014
	ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx);
2015
	if (ret < 0) {
2016 2017
		/* Fallthrough and commit/free transaction. */
		ret = 1;
2018
	}
C
Chris Mason 已提交
2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029

	/* we've logged all the items and now have a consistent
	 * version of the file in the log.  It is possible that
	 * someone will come in and modify the file, but that's
	 * fine because the log is consistent on disk, and we
	 * have references to all of the file's extents
	 *
	 * It is possible that someone will come in and log the
	 * file again, but that will end up using the synchronization
	 * inside btrfs_sync_log to keep things safe.
	 */
2030
	mutex_unlock(&inode->i_mutex);
C
Chris Mason 已提交
2031

2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050
	/*
	 * If any of the ordered extents had an error, just return it to user
	 * space, so that the application knows some writes didn't succeed and
	 * can take proper action (retry for e.g.). Blindly committing the
	 * transaction in this case, would fool userspace that everything was
	 * successful. And we also want to make sure our log doesn't contain
	 * file extent items pointing to extents that weren't fully written to -
	 * just like in the non fast fsync path, where we check for the ordered
	 * operation's error flag before writing to the log tree and return -EIO
	 * if any of them had this flag set (btrfs_wait_ordered_range) -
	 * therefore we need to check for errors in the ordered operations,
	 * which are indicated by ctx.io_err.
	 */
	if (ctx.io_err) {
		btrfs_end_transaction(trans, root);
		ret = ctx.io_err;
		goto out;
	}

2051
	if (ret != BTRFS_NO_LOG_SYNC) {
2052
		if (!ret) {
2053
			ret = btrfs_sync_log(trans, root, &ctx);
2054
			if (!ret) {
2055
				ret = btrfs_end_transaction(trans, root);
2056
				goto out;
2057
			}
2058
		}
2059 2060 2061
		if (!full_sync) {
			ret = btrfs_wait_ordered_range(inode, start,
						       end - start + 1);
2062 2063
			if (ret) {
				btrfs_end_transaction(trans, root);
2064
				goto out;
2065
			}
2066 2067
		}
		ret = btrfs_commit_transaction(trans, root);
2068 2069
	} else {
		ret = btrfs_end_transaction(trans, root);
2070
	}
C
Chris Mason 已提交
2071
out:
2072
	return ret > 0 ? -EIO : ret;
C
Chris Mason 已提交
2073 2074
}

2075
static const struct vm_operations_struct btrfs_file_vm_ops = {
2076
	.fault		= filemap_fault,
2077
	.map_pages	= filemap_map_pages,
C
Chris Mason 已提交
2078 2079 2080 2081 2082
	.page_mkwrite	= btrfs_page_mkwrite,
};

static int btrfs_file_mmap(struct file	*filp, struct vm_area_struct *vma)
{
M
Miao Xie 已提交
2083 2084 2085 2086 2087
	struct address_space *mapping = filp->f_mapping;

	if (!mapping->a_ops->readpage)
		return -ENOEXEC;

C
Chris Mason 已提交
2088
	file_accessed(filp);
M
Miao Xie 已提交
2089 2090
	vma->vm_ops = &btrfs_file_vm_ops;

C
Chris Mason 已提交
2091 2092 2093
	return 0;
}

J
Josef Bacik 已提交
2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133
static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
			  int slot, u64 start, u64 end)
{
	struct btrfs_file_extent_item *fi;
	struct btrfs_key key;

	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
		return 0;

	btrfs_item_key_to_cpu(leaf, &key, slot);
	if (key.objectid != btrfs_ino(inode) ||
	    key.type != BTRFS_EXTENT_DATA_KEY)
		return 0;

	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);

	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
		return 0;

	if (btrfs_file_extent_disk_bytenr(leaf, fi))
		return 0;

	if (key.offset == end)
		return 1;
	if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
		return 1;
	return 0;
}

static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
		      struct btrfs_path *path, u64 offset, u64 end)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct extent_buffer *leaf;
	struct btrfs_file_extent_item *fi;
	struct extent_map *hole_em;
	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
	struct btrfs_key key;
	int ret;

2134 2135 2136
	if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
		goto out;

J
Josef Bacik 已提交
2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161
	key.objectid = btrfs_ino(inode);
	key.type = BTRFS_EXTENT_DATA_KEY;
	key.offset = offset;

	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
	if (ret < 0)
		return ret;
	BUG_ON(!ret);

	leaf = path->nodes[0];
	if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
		u64 num_bytes;

		path->slots[0]--;
		fi = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_file_extent_item);
		num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
			end - offset;
		btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
		btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
		btrfs_set_file_extent_offset(leaf, fi, 0);
		btrfs_mark_buffer_dirty(leaf);
		goto out;
	}

2162
	if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
J
Josef Bacik 已提交
2163 2164 2165
		u64 num_bytes;

		key.offset = offset;
2166
		btrfs_set_item_key_safe(root->fs_info, path, &key);
J
Josef Bacik 已提交
2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195
		fi = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_file_extent_item);
		num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
			offset;
		btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
		btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
		btrfs_set_file_extent_offset(leaf, fi, 0);
		btrfs_mark_buffer_dirty(leaf);
		goto out;
	}
	btrfs_release_path(path);

	ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
				       0, 0, end - offset, 0, end - offset,
				       0, 0, 0);
	if (ret)
		return ret;

out:
	btrfs_release_path(path);

	hole_em = alloc_extent_map();
	if (!hole_em) {
		btrfs_drop_extent_cache(inode, offset, end - 1, 0);
		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
			&BTRFS_I(inode)->runtime_flags);
	} else {
		hole_em->start = offset;
		hole_em->len = end - offset;
J
Josef Bacik 已提交
2196
		hole_em->ram_bytes = hole_em->len;
J
Josef Bacik 已提交
2197 2198 2199 2200
		hole_em->orig_start = offset;

		hole_em->block_start = EXTENT_MAP_HOLE;
		hole_em->block_len = 0;
2201
		hole_em->orig_block_len = 0;
J
Josef Bacik 已提交
2202 2203 2204 2205 2206 2207 2208
		hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
		hole_em->compress_type = BTRFS_COMPRESS_NONE;
		hole_em->generation = trans->transid;

		do {
			btrfs_drop_extent_cache(inode, offset, end - 1, 0);
			write_lock(&em_tree->lock);
J
Josef Bacik 已提交
2209
			ret = add_extent_mapping(em_tree, hole_em, 1);
J
Josef Bacik 已提交
2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220
			write_unlock(&em_tree->lock);
		} while (ret == -EEXIST);
		free_extent_map(hole_em);
		if (ret)
			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
				&BTRFS_I(inode)->runtime_flags);
	}

	return 0;
}

2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251
/*
 * Find a hole extent on given inode and change start/len to the end of hole
 * extent.(hole/vacuum extent whose em->start <= start &&
 *	   em->start + em->len > start)
 * When a hole extent is found, return 1 and modify start/len.
 */
static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
{
	struct extent_map *em;
	int ret = 0;

	em = btrfs_get_extent(inode, NULL, 0, *start, *len, 0);
	if (IS_ERR_OR_NULL(em)) {
		if (!em)
			ret = -ENOMEM;
		else
			ret = PTR_ERR(em);
		return ret;
	}

	/* Hole or vacuum extent(only exists in no-hole mode) */
	if (em->block_start == EXTENT_MAP_HOLE) {
		ret = 1;
		*len = em->start + em->len > *start + *len ?
		       0 : *start + *len - em->start - em->len;
		*start = em->start + em->len;
	}
	free_extent_map(em);
	return ret;
}

J
Josef Bacik 已提交
2252 2253 2254 2255 2256 2257 2258
static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct extent_state *cached_state = NULL;
	struct btrfs_path *path;
	struct btrfs_block_rsv *rsv;
	struct btrfs_trans_handle *trans;
2259 2260 2261 2262 2263 2264
	u64 lockstart;
	u64 lockend;
	u64 tail_start;
	u64 tail_len;
	u64 orig_start = offset;
	u64 cur_offset;
J
Josef Bacik 已提交
2265 2266 2267 2268
	u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
	u64 drop_end;
	int ret = 0;
	int err = 0;
2269
	int rsv_count;
2270
	bool same_page;
2271
	bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2272
	u64 ino_size;
2273 2274
	bool truncated_page = false;
	bool updated_inode = false;
J
Josef Bacik 已提交
2275

2276 2277 2278
	ret = btrfs_wait_ordered_range(inode, offset, len);
	if (ret)
		return ret;
J
Josef Bacik 已提交
2279 2280

	mutex_lock(&inode->i_mutex);
2281
	ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2282 2283 2284 2285 2286 2287 2288 2289 2290
	ret = find_first_non_hole(inode, &offset, &len);
	if (ret < 0)
		goto out_only_mutex;
	if (ret && !len) {
		/* Already in a large hole */
		ret = 0;
		goto out_only_mutex;
	}

2291
	lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2292 2293 2294 2295 2296
	lockend = round_down(offset + len,
			     BTRFS_I(inode)->root->sectorsize) - 1;
	same_page = ((offset >> PAGE_CACHE_SHIFT) ==
		    ((offset + len - 1) >> PAGE_CACHE_SHIFT));

2297 2298 2299 2300
	/*
	 * We needn't truncate any page which is beyond the end of the file
	 * because we are sure there is no data there.
	 */
J
Josef Bacik 已提交
2301 2302 2303 2304 2305
	/*
	 * Only do this if we are in the same page and we aren't doing the
	 * entire page.
	 */
	if (same_page && len < PAGE_CACHE_SIZE) {
2306 2307
		if (offset < ino_size) {
			truncated_page = true;
2308
			ret = btrfs_truncate_page(inode, offset, len, 0);
2309 2310 2311
		} else {
			ret = 0;
		}
2312
		goto out_only_mutex;
J
Josef Bacik 已提交
2313 2314 2315
	}

	/* zero back part of the first page */
2316
	if (offset < ino_size) {
2317
		truncated_page = true;
2318 2319 2320 2321 2322
		ret = btrfs_truncate_page(inode, offset, 0, 0);
		if (ret) {
			mutex_unlock(&inode->i_mutex);
			return ret;
		}
J
Josef Bacik 已提交
2323 2324
	}

2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352
	/* Check the aligned pages after the first unaligned page,
	 * if offset != orig_start, which means the first unaligned page
	 * including serveral following pages are already in holes,
	 * the extra check can be skipped */
	if (offset == orig_start) {
		/* after truncate page, check hole again */
		len = offset + len - lockstart;
		offset = lockstart;
		ret = find_first_non_hole(inode, &offset, &len);
		if (ret < 0)
			goto out_only_mutex;
		if (ret && !len) {
			ret = 0;
			goto out_only_mutex;
		}
		lockstart = offset;
	}

	/* Check the tail unaligned part is in a hole */
	tail_start = lockend + 1;
	tail_len = offset + len - tail_start;
	if (tail_len) {
		ret = find_first_non_hole(inode, &tail_start, &tail_len);
		if (unlikely(ret < 0))
			goto out_only_mutex;
		if (!ret) {
			/* zero the front end of the last page */
			if (tail_start + tail_len < ino_size) {
2353
				truncated_page = true;
2354 2355 2356 2357
				ret = btrfs_truncate_page(inode,
						tail_start + tail_len, 0, 1);
				if (ret)
					goto out_only_mutex;
2358
			}
M
Miao Xie 已提交
2359
		}
J
Josef Bacik 已提交
2360 2361 2362
	}

	if (lockend < lockstart) {
2363 2364
		ret = 0;
		goto out_only_mutex;
J
Josef Bacik 已提交
2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381
	}

	while (1) {
		struct btrfs_ordered_extent *ordered;

		truncate_pagecache_range(inode, lockstart, lockend);

		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
				 0, &cached_state);
		ordered = btrfs_lookup_first_ordered_extent(inode, lockend);

		/*
		 * We need to make sure we have no ordered extents in this range
		 * and nobody raced in and read a page in this range, if we did
		 * we need to try again.
		 */
		if ((!ordered ||
2382
		    (ordered->file_offset + ordered->len <= lockstart ||
J
Josef Bacik 已提交
2383
		     ordered->file_offset > lockend)) &&
2384
		     !btrfs_page_exists_in_range(inode, lockstart, lockend)) {
J
Josef Bacik 已提交
2385 2386 2387 2388 2389 2390 2391 2392
			if (ordered)
				btrfs_put_ordered_extent(ordered);
			break;
		}
		if (ordered)
			btrfs_put_ordered_extent(ordered);
		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
				     lockend, &cached_state, GFP_NOFS);
2393 2394 2395 2396 2397 2398
		ret = btrfs_wait_ordered_range(inode, lockstart,
					       lockend - lockstart + 1);
		if (ret) {
			mutex_unlock(&inode->i_mutex);
			return ret;
		}
J
Josef Bacik 已提交
2399 2400 2401 2402 2403 2404 2405 2406
	}

	path = btrfs_alloc_path();
	if (!path) {
		ret = -ENOMEM;
		goto out;
	}

2407
	rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
J
Josef Bacik 已提交
2408 2409 2410 2411 2412 2413 2414 2415 2416 2417
	if (!rsv) {
		ret = -ENOMEM;
		goto out_free;
	}
	rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
	rsv->failfast = 1;

	/*
	 * 1 - update the inode
	 * 1 - removing the extents in the range
2418
	 * 1 - adding the hole extent if no_holes isn't set
J
Josef Bacik 已提交
2419
	 */
2420 2421
	rsv_count = no_holes ? 2 : 3;
	trans = btrfs_start_transaction(root, rsv_count);
J
Josef Bacik 已提交
2422 2423 2424 2425 2426 2427 2428 2429 2430 2431
	if (IS_ERR(trans)) {
		err = PTR_ERR(trans);
		goto out_free;
	}

	ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
				      min_size);
	BUG_ON(ret);
	trans->block_rsv = rsv;

2432 2433
	cur_offset = lockstart;
	len = lockend - cur_offset;
J
Josef Bacik 已提交
2434 2435 2436
	while (cur_offset < lockend) {
		ret = __btrfs_drop_extents(trans, root, inode, path,
					   cur_offset, lockend + 1,
2437
					   &drop_end, 1, 0, 0, NULL);
J
Josef Bacik 已提交
2438 2439 2440 2441 2442
		if (ret != -ENOSPC)
			break;

		trans->block_rsv = &root->fs_info->trans_block_rsv;

2443 2444 2445 2446 2447 2448 2449
		if (cur_offset < ino_size) {
			ret = fill_holes(trans, inode, path, cur_offset,
					 drop_end);
			if (ret) {
				err = ret;
				break;
			}
J
Josef Bacik 已提交
2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460
		}

		cur_offset = drop_end;

		ret = btrfs_update_inode(trans, root, inode);
		if (ret) {
			err = ret;
			break;
		}

		btrfs_end_transaction(trans, root);
2461
		btrfs_btree_balance_dirty(root);
J
Josef Bacik 已提交
2462

2463
		trans = btrfs_start_transaction(root, rsv_count);
J
Josef Bacik 已提交
2464 2465 2466 2467 2468 2469 2470 2471 2472 2473
		if (IS_ERR(trans)) {
			ret = PTR_ERR(trans);
			trans = NULL;
			break;
		}

		ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
					      rsv, min_size);
		BUG_ON(ret);	/* shouldn't happen */
		trans->block_rsv = rsv;
2474 2475 2476 2477 2478 2479 2480 2481

		ret = find_first_non_hole(inode, &cur_offset, &len);
		if (unlikely(ret < 0))
			break;
		if (ret && !len) {
			ret = 0;
			break;
		}
J
Josef Bacik 已提交
2482 2483 2484 2485 2486 2487 2488 2489
	}

	if (ret) {
		err = ret;
		goto out_trans;
	}

	trans->block_rsv = &root->fs_info->trans_block_rsv;
2490 2491 2492 2493 2494 2495
	/*
	 * Don't insert file hole extent item if it's for a range beyond eof
	 * (because it's useless) or if it represents a 0 bytes range (when
	 * cur_offset == drop_end).
	 */
	if (cur_offset < ino_size && cur_offset < drop_end) {
2496 2497 2498 2499 2500
		ret = fill_holes(trans, inode, path, cur_offset, drop_end);
		if (ret) {
			err = ret;
			goto out_trans;
		}
J
Josef Bacik 已提交
2501 2502 2503 2504 2505 2506
	}

out_trans:
	if (!trans)
		goto out_free;

2507 2508 2509
	inode_inc_iversion(inode);
	inode->i_mtime = inode->i_ctime = CURRENT_TIME;

J
Josef Bacik 已提交
2510 2511
	trans->block_rsv = &root->fs_info->trans_block_rsv;
	ret = btrfs_update_inode(trans, root, inode);
2512
	updated_inode = true;
J
Josef Bacik 已提交
2513
	btrfs_end_transaction(trans, root);
2514
	btrfs_btree_balance_dirty(root);
J
Josef Bacik 已提交
2515 2516 2517 2518 2519 2520
out_free:
	btrfs_free_path(path);
	btrfs_free_block_rsv(root, rsv);
out:
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
			     &cached_state, GFP_NOFS);
2521
out_only_mutex:
2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537
	if (!updated_inode && truncated_page && !ret && !err) {
		/*
		 * If we only end up zeroing part of a page, we still need to
		 * update the inode item, so that all the time fields are
		 * updated as well as the necessary btrfs inode in memory fields
		 * for detecting, at fsync time, if the inode isn't yet in the
		 * log tree or it's there but not up to date.
		 */
		trans = btrfs_start_transaction(root, 1);
		if (IS_ERR(trans)) {
			err = PTR_ERR(trans);
		} else {
			err = btrfs_update_inode(trans, root, inode);
			ret = btrfs_end_transaction(trans, root);
		}
	}
J
Josef Bacik 已提交
2538 2539 2540 2541 2542 2543
	mutex_unlock(&inode->i_mutex);
	if (ret && !err)
		err = ret;
	return err;
}

2544 2545 2546
static long btrfs_fallocate(struct file *file, int mode,
			    loff_t offset, loff_t len)
{
A
Al Viro 已提交
2547
	struct inode *inode = file_inode(file);
2548 2549 2550 2551 2552 2553 2554 2555
	struct extent_state *cached_state = NULL;
	u64 cur_offset;
	u64 last_byte;
	u64 alloc_start;
	u64 alloc_end;
	u64 alloc_hint = 0;
	u64 locked_end;
	struct extent_map *em;
2556
	int blocksize = BTRFS_I(inode)->root->sectorsize;
2557 2558
	int ret;

2559 2560
	alloc_start = round_down(offset, blocksize);
	alloc_end = round_up(offset + len, blocksize);
2561

J
Josef Bacik 已提交
2562 2563
	/* Make sure we aren't being give some crap mode */
	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2564 2565
		return -EOPNOTSUPP;

J
Josef Bacik 已提交
2566 2567 2568
	if (mode & FALLOC_FL_PUNCH_HOLE)
		return btrfs_punch_hole(inode, offset, len);

2569 2570 2571 2572
	/*
	 * Make sure we have enough space before we do the
	 * allocation.
	 */
2573
	ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start, alloc_end - alloc_start);
2574 2575 2576
	if (ret)
		return ret;

2577 2578 2579 2580 2581 2582
	mutex_lock(&inode->i_mutex);
	ret = inode_newsize_ok(inode, alloc_end);
	if (ret)
		goto out;

	if (alloc_start > inode->i_size) {
2583 2584
		ret = btrfs_cont_expand(inode, i_size_read(inode),
					alloc_start);
2585 2586
		if (ret)
			goto out;
2587
	} else if (offset + len > inode->i_size) {
2588 2589 2590 2591 2592 2593 2594 2595
		/*
		 * If we are fallocating from the end of the file onward we
		 * need to zero out the end of the page if i_size lands in the
		 * middle of a page.
		 */
		ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
		if (ret)
			goto out;
2596 2597
	}

2598 2599 2600 2601
	/*
	 * wait for ordered IO before we have any locks.  We'll loop again
	 * below with the locks held.
	 */
2602 2603 2604 2605
	ret = btrfs_wait_ordered_range(inode, alloc_start,
				       alloc_end - alloc_start);
	if (ret)
		goto out;
2606

2607 2608 2609 2610 2611 2612 2613 2614
	locked_end = alloc_end - 1;
	while (1) {
		struct btrfs_ordered_extent *ordered;

		/* the extent lock is ordered inside the running
		 * transaction
		 */
		lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2615
				 locked_end, 0, &cached_state);
2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628
		ordered = btrfs_lookup_first_ordered_extent(inode,
							    alloc_end - 1);
		if (ordered &&
		    ordered->file_offset + ordered->len > alloc_start &&
		    ordered->file_offset < alloc_end) {
			btrfs_put_ordered_extent(ordered);
			unlock_extent_cached(&BTRFS_I(inode)->io_tree,
					     alloc_start, locked_end,
					     &cached_state, GFP_NOFS);
			/*
			 * we can't wait on the range with the transaction
			 * running or with the extent lock held
			 */
2629 2630 2631 2632
			ret = btrfs_wait_ordered_range(inode, alloc_start,
						       alloc_end - alloc_start);
			if (ret)
				goto out;
2633 2634 2635 2636 2637 2638 2639 2640 2641
		} else {
			if (ordered)
				btrfs_put_ordered_extent(ordered);
			break;
		}
	}

	cur_offset = alloc_start;
	while (1) {
2642 2643
		u64 actual_end;

2644 2645
		em = btrfs_get_extent(inode, NULL, 0, cur_offset,
				      alloc_end - cur_offset, 0);
2646 2647 2648 2649 2650 2651 2652
		if (IS_ERR_OR_NULL(em)) {
			if (!em)
				ret = -ENOMEM;
			else
				ret = PTR_ERR(em);
			break;
		}
2653
		last_byte = min(extent_map_end(em), alloc_end);
2654
		actual_end = min_t(u64, extent_map_end(em), offset + len);
2655
		last_byte = ALIGN(last_byte, blocksize);
2656

2657 2658 2659 2660 2661 2662 2663 2664
		if (em->block_start == EXTENT_MAP_HOLE ||
		    (cur_offset >= inode->i_size &&
		     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
			ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
							last_byte - cur_offset,
							1 << inode->i_blkbits,
							offset + len,
							&alloc_hint);
2665 2666
		} else if (actual_end > inode->i_size &&
			   !(mode & FALLOC_FL_KEEP_SIZE)) {
2667
			struct btrfs_trans_handle *trans;
2668
			struct btrfs_root *root = BTRFS_I(inode)->root;
2669

2670 2671 2672
			/*
			 * We didn't need to allocate any more space, but we
			 * still extended the size of the file so we need to
2673
			 * update i_size and the inode item.
2674
			 */
2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689
			trans = btrfs_start_transaction(root, 1);
			if (IS_ERR(trans)) {
				ret = PTR_ERR(trans);
			} else {
				inode->i_ctime = CURRENT_TIME;
				i_size_write(inode, actual_end);
				btrfs_ordered_update_i_size(inode, actual_end,
							    NULL);
				ret = btrfs_update_inode(trans, root, inode);
				if (ret)
					btrfs_end_transaction(trans, root);
				else
					ret = btrfs_end_transaction(trans,
								    root);
			}
2690 2691
		}
		free_extent_map(em);
2692 2693
		if (ret < 0)
			break;
2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704

		cur_offset = last_byte;
		if (cur_offset >= alloc_end) {
			ret = 0;
			break;
		}
	}
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
			     &cached_state, GFP_NOFS);
out:
	mutex_unlock(&inode->i_mutex);
2705
	/* Let go of our reservation. */
2706
	btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2707 2708 2709
	return ret;
}

2710
static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
J
Josef Bacik 已提交
2711 2712
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
2713
	struct extent_map *em = NULL;
J
Josef Bacik 已提交
2714
	struct extent_state *cached_state = NULL;
L
Liu Bo 已提交
2715 2716 2717 2718
	u64 lockstart;
	u64 lockend;
	u64 start;
	u64 len;
J
Josef Bacik 已提交
2719 2720
	int ret = 0;

L
Liu Bo 已提交
2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731
	if (inode->i_size == 0)
		return -ENXIO;

	/*
	 * *offset can be negative, in this case we start finding DATA/HOLE from
	 * the very start of the file.
	 */
	start = max_t(loff_t, 0, *offset);

	lockstart = round_down(start, root->sectorsize);
	lockend = round_up(i_size_read(inode), root->sectorsize);
J
Josef Bacik 已提交
2732 2733
	if (lockend <= lockstart)
		lockend = lockstart + root->sectorsize;
L
Liu Bo 已提交
2734
	lockend--;
J
Josef Bacik 已提交
2735 2736 2737
	len = lockend - lockstart + 1;

	lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2738
			 &cached_state);
J
Josef Bacik 已提交
2739

2740
	while (start < inode->i_size) {
J
Josef Bacik 已提交
2741 2742
		em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
		if (IS_ERR(em)) {
2743
			ret = PTR_ERR(em);
2744
			em = NULL;
J
Josef Bacik 已提交
2745 2746 2747
			break;
		}

2748 2749 2750 2751 2752 2753 2754 2755
		if (whence == SEEK_HOLE &&
		    (em->block_start == EXTENT_MAP_HOLE ||
		     test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
			break;
		else if (whence == SEEK_DATA &&
			   (em->block_start != EXTENT_MAP_HOLE &&
			    !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
			break;
J
Josef Bacik 已提交
2756 2757 2758

		start = em->start + em->len;
		free_extent_map(em);
2759
		em = NULL;
J
Josef Bacik 已提交
2760 2761
		cond_resched();
	}
2762 2763 2764 2765 2766 2767 2768
	free_extent_map(em);
	if (!ret) {
		if (whence == SEEK_DATA && start >= inode->i_size)
			ret = -ENXIO;
		else
			*offset = min_t(loff_t, start, inode->i_size);
	}
J
Josef Bacik 已提交
2769 2770 2771 2772 2773
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
			     &cached_state, GFP_NOFS);
	return ret;
}

2774
static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
J
Josef Bacik 已提交
2775 2776 2777 2778 2779
{
	struct inode *inode = file->f_mapping->host;
	int ret;

	mutex_lock(&inode->i_mutex);
2780
	switch (whence) {
J
Josef Bacik 已提交
2781 2782
	case SEEK_END:
	case SEEK_CUR:
2783
		offset = generic_file_llseek(file, offset, whence);
J
Josef Bacik 已提交
2784 2785 2786
		goto out;
	case SEEK_DATA:
	case SEEK_HOLE:
2787 2788 2789 2790 2791
		if (offset >= i_size_read(inode)) {
			mutex_unlock(&inode->i_mutex);
			return -ENXIO;
		}

2792
		ret = find_desired_extent(inode, &offset, whence);
J
Josef Bacik 已提交
2793 2794 2795 2796 2797 2798
		if (ret) {
			mutex_unlock(&inode->i_mutex);
			return ret;
		}
	}

J
Jie Liu 已提交
2799
	offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
J
Josef Bacik 已提交
2800 2801 2802 2803 2804
out:
	mutex_unlock(&inode->i_mutex);
	return offset;
}

2805
const struct file_operations btrfs_file_operations = {
J
Josef Bacik 已提交
2806
	.llseek		= btrfs_file_llseek,
2807
	.read_iter      = generic_file_read_iter,
C
Chris Mason 已提交
2808
	.splice_read	= generic_file_splice_read,
A
Al Viro 已提交
2809
	.write_iter	= btrfs_file_write_iter,
C
Chris Mason 已提交
2810
	.mmap		= btrfs_file_mmap,
C
Chris Mason 已提交
2811
	.open		= generic_file_open,
2812
	.release	= btrfs_release_file,
C
Chris Mason 已提交
2813
	.fsync		= btrfs_sync_file,
2814
	.fallocate	= btrfs_fallocate,
C
Christoph Hellwig 已提交
2815
	.unlocked_ioctl	= btrfs_ioctl,
C
Chris Mason 已提交
2816
#ifdef CONFIG_COMPAT
C
Christoph Hellwig 已提交
2817
	.compat_ioctl	= btrfs_ioctl,
C
Chris Mason 已提交
2818 2819
#endif
};
2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837

void btrfs_auto_defrag_exit(void)
{
	if (btrfs_inode_defrag_cachep)
		kmem_cache_destroy(btrfs_inode_defrag_cachep);
}

int btrfs_auto_defrag_init(void)
{
	btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
					sizeof(struct inode_defrag), 0,
					SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
					NULL);
	if (!btrfs_inode_defrag_cachep)
		return -ENOMEM;

	return 0;
}
2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863

int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
{
	int ret;

	/*
	 * So with compression we will find and lock a dirty page and clear the
	 * first one as dirty, setup an async extent, and immediately return
	 * with the entire range locked but with nobody actually marked with
	 * writeback.  So we can't just filemap_write_and_wait_range() and
	 * expect it to work since it will just kick off a thread to do the
	 * actual work.  So we need to call filemap_fdatawrite_range _again_
	 * since it will wait on the page lock, which won't be unlocked until
	 * after the pages have been marked as writeback and so we're good to go
	 * from there.  We have to do this otherwise we'll miss the ordered
	 * extents and that results in badness.  Please Josef, do not think you
	 * know better and pull this out at some point in the future, it is
	 * right and you are wrong.
	 */
	ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
	if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
			     &BTRFS_I(inode)->runtime_flags))
		ret = filemap_fdatawrite_range(inode->i_mapping, start, end);

	return ret;
}